Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Simulation and Compensation Methods for EUV Lithography Masks with Buried Defects
by
Chris Heinz Clifford
A dissertation submitted in partial satisfaction of the requirements for the degree of
Doctor of Philosophy
in
Engineering - Electrical Engineering and Computer Sciences
in the
Graduate Division
of the
University of California, Berkeley
Committee in charge:
Professor Andrew R. Neureuther
Professor Costas J. Spanos
Professor Xiang Zhang
Spring 2010
Simulation and Compensation Methods for EUV Lithography Masks with Buried Defects
Copyright © 2010
by
Chris Heinz Clifford
All rights reserved
1
Abstract
Simulation and Compensation Methods for EUV Lithography Masks with Buried Defects
by
Chris Heinz Clifford
Doctor of Philosophy in Engineering - Electrical Engineering and Computer Sciences
University of California, Berkeley
Professor Andrew R. Neureuther, Chair
This dissertation describes the development and application of a new simulator,
RADICAL, which can accurately simulate the electromagnetic interaction in extreme ultraviolet
(EUV) lithography at a wavelength of 13.5nm between the mask absorber features and a buried
multilayer defect three orders of magnitude faster than rigorous methods. RADICAL achieves
this performance by using simulation and modeling methods designed specifically for the
individual EUV mask components simulated. The nonplanar nature of a multilayer coated
buried defect can be simulated using a ray tracing method developed by Michael Lam, or the
faster advanced single surface approximation (SSA) for shorter defects with more uniform layers
below the mask surface, and the absorber is modeled using a propagated thin mask model which
efficiently accounts for the thickness of the absorber material. The multilayer and absorber
simulation results are linked by a Fourier transform which converts the electric field output by
one simulator into a set of plane waves for the next.
As a new form of projection lithography technology, EUV is fundamentally different than
current technologies and therefore requires new methods for mask analysis, inspection and
compensation. At the 13.5nm wavelength, all materials have a refractive index of around unity
and are absorptive. This means that EUV masks must be reflective multilayers. Understanding
the electromagnetic response of these new masks, specifically the effects of multilayer defects
requires new simulation methods, such as RADICAL.
The accuracy and speed of RADICAL has been verified by comparisons with rigorous
finite difference time domain (FDTD) simulations, rigorous waveguide method simulations, and
actinic inspection experiments provided by Lawrence Berkeley National Laboratory (LBNL) and
Intel. RADICAL matches the critical dimension (CD) predicted by FDTD within 1nm for
defects up to 2.5nm tall on the multilayer surface. It matches actinic inspection results, within
the error of the experiment, for defects up to 6.5nm tall on the surface. This accuracy is
acceptable because the EUV defects expected in production lithography are expected be 2nm tall
or less. RADICAL is typically about 1,000 times faster than FDTD for a single simulation of a
two-dimensional absorber pattern. But, RADICAL’s modular design allows the re-use of
multilayer simulation results so subsequent simulations of different patterns over the same defect
geometry take a small fraction of the time of the first simulation.
RADICAL has been applied to many important issues in EUV lithography. It was used to
advance the understanding of isolated defects by showing that they are primarily phase defects,
because they cause an inversion in aerial image intensity through focus, but cannot simply be
2
modeled by a thin mask defect with a uniform phase. This means EUV buried defects must be
treated differently than the phase defects encountered in conventional optical lithography masks.
Experimental images of isolated defects were used to extract large aberrations from the
state of the art Actinic Inspection Tool (AIT) at LBNL. This novel method for aberration
extraction required only the comparison of the center intensity of an isolated defect image
through focus between RADICAL and the AIT images. This work resulted in a deeper
understanding of the AIT tool, which led to improvements of the tool by the team at LBNL,
which have benefited the entire industry.
The critical issue of the printability of buried defects near features and its dependence on
illumination in future production tools is addressed by a thorough investigation of the resulting
aerial images and CD change for defects near 22nm and 16nm features. The effect of the
position, size and shape of the defect is explored. The printability of a buried defect is very
dependent on its position relative to the absorber features and the worst case position depends on
the defect size. Also, defects only a few nanometers tall that are covered by the absorber can still
cause an unacceptable CD change. The effects of illumination are also investigated to show that
the improved image slope produced by advanced off-axis illuminations reduces the printability
of buried defects in focus, but through focus the area affected by a buried defect is much larger
for dipole and annular illuminations than it is for tophat.
The final topic of this dissertation is defect compensation. Two methods are proposed to
reduce the effects of a buried defect by adjusting the absorber pattern. The first employs pre-
calculated design curves to prescribe absorber modifications based only on the CD change
caused by the defect. This method successfully compensates for the defect in focus, but the
defect still causes a CD change through focus. To reduce the effect of the defect through focus,
absorber is used to cover the defect and block the out of phase light from the defect. This
covering produces some improvement in through focus printability and works best for defects
which have a narrower surface geometry after smoothing.
i
Dedicated to my Mom, Dad, Clare and Jaime for always having confidence in me
ii
Acknowledgements Looking back over my graduate school career, the most important person to acknowledge
is my advisor Andy Neureuther. He provided me with everything you could ask for from an
advisor. He was always there when I needed help with anything from a difficult research
problem to a confusing bureaucratic procedure. He made sure that I was doing the work that
needed to get done, but understood that there was a lot more to graduate school than publishing
papers.
Over the course of my graduate career I was lucky enough to travel all over the world to
present my research and listen to others present their research. I was able to do this because of
the generous support of Intel, who funded my entire graduate school career. Ted Liang at Intel
was not just sponsor, but a technical mentor and friend.
When I was starting this research project I knew very little about lithography, which made
doing new research difficult. Luckily, Michael Lam spent many hours explaining his Ph.D. work
and broader lithography concepts to me. His help saved me months of frustration and I would
not have been able to do a lot of the work in this dissertation without it.
I enjoyed working with and learning from the other students in Andy’s group who
graduated while I was in Berkeley: Greg McIntyre, Wojciech Poppe, and Dan Ceperley. Dan
was an especially good friend and I would not have been able to do any of the FDTD simulations
in this dissertation without him. He was also the founder of Photobears, the student group for
SPIE and OSA at Berkeley.
There are four students who were in Andy’s group with me for my entire time in graduate
school: Juliet Rubinstein, Eric Chin, Lynn Wang, and Marshal Miller. I did not collaborate on
research projects with them very often, but I was lucky to have all of them for help on everything
from studying for the preliminary exam to finding cheap transatlantic flights.
In 2008 we were lucky to have Kenji Yamazoe, a visiting scholar from Canon, join our
group. He is a kind and fun person who is also one of the top experts in the world on
computational methods for calculating a lithographic image. His help was invaluable to me,
especially as I prepared for my qualifying exam.
The University of California – Berkeley is located just down the hill from Lawrence
Berkeley National Lab (LBNL). LBNL has many top researchers in EUV lithography. I was
lucky enough to work closely with two of them, Ken Goldberg and Iacopo Mochi. A large
portion of this dissertation would not have been possible without their help.
Throughout my time in graduate school I could always count on the support of my family
and girlfriend Jaime. My family was always a phone call away if I needed advice or someone to
talk to. Being with Jaime, I always looked forward to coming home at the end of the day. Most
of the fun I had as a graduate student was with her. Everywhere from the pool at the Laurels in
San Jose to the top of the Eiffel Tower, we had fun together and I am lucky to have her in my
life.
iii
Table of Contents 1 Introduction .................................................................................................................. 1
1.1 Motivation ............................................................................................................. 1
1.2 Dissertation Content and Contributions ................................................................ 2
2 EUV Lithography ......................................................................................................... 5
2.1 Lithography for Semiconductor Manufacturing ................................................... 5
2.1.1 Illumination ...................................................................................................... 6
2.1.2 Photomask ........................................................................................................ 7
2.1.3 Projection Optics .............................................................................................. 7
2.2 Lithography with Extreme Ultraviolet Light ........................................................ 8
2.2.1 Illumination for EUV Lithography .................................................................. 8
2.2.2 Photomask for EUV Lithography .................................................................... 9
2.2.3 Projection Optics for EUV Lithography ........................................................ 10
2.3 Current issues in EUV Lithography Development ............................................. 10
2.3.1 Source ............................................................................................................ 10
2.3.2 Photoresist ...................................................................................................... 10
2.3.3 Mask ............................................................................................................... 11
2.4 Existing EUV Lithography Tools ....................................................................... 12
2.4.1 Full Field Scanners ........................................................................................ 12
2.4.2 Micro Exposure Tools.................................................................................... 12
2.4.3 Actinic Inspection Tools ................................................................................ 12
2.4.4 Programmed Defect Mask (PDM) ................................................................. 13
2.5 Summary ............................................................................................................. 14
3 Fast Simulation Method for EUV Masks with Buried Defects .................................. 15
3.1 History of EUV Mask Simulation ....................................................................... 15
3.2 RADICAL ........................................................................................................... 16
3.2.1 Multilayer Simulation .................................................................................... 17
3.2.2 Absorber Simulation ...................................................................................... 18
3.2.3 Accuracy of RADICAL ................................................................................. 19
3.2.4 Computational Requirements of RADICAL ................................................. 23
3.3 Summary ............................................................................................................. 25
4 Printability of Isolated Defects ................................................................................... 26
4.1 Phase Nature of Defects ...................................................................................... 26
4.2 Simple Model for Isolated Defects...................................................................... 27
4.2.1 Introduction to Smoothing ............................................................................. 27
4.2.2 Analysis of Smoothing ................................................................................... 29
4.2.3 Algebraic Isolated Defect Model ................................................................... 33
iv
4.2.4 Tuned Single Surface Approximation ........................................................... 34
4.3 Effects of Illumination ........................................................................................ 36
4.4 Mask blank inspection ......................................................................................... 37
4.5 Isolated Defect Experiments ............................................................................... 38
4.6 Summary ............................................................................................................. 40
5 Using Isolated Buried Defects to Extract Aberrations ............................................... 41
5.1 Astigmatism ........................................................................................................ 42
5.2 Spherical .............................................................................................................. 43
5.3 Coma ................................................................................................................... 44
5.4 Limitations of Method ......................................................................................... 45
5.5 Summary ............................................................................................................. 45
6 Printability of Defects Near Features ......................................................................... 46
6.1 Effects of Defect Size .......................................................................................... 46
6.2 Effects of Defect Position ................................................................................... 47
6.3 Effects of Focus ................................................................................................... 48
6.4 Effects of Illumination ........................................................................................ 49
6.5 Experimental Images of Defects Near Features .................................................. 51
6.5.1 Printability as a Function of Defect Size ....................................................... 51
6.5.2 Printability of Covered Defects ..................................................................... 54
6.5.3 Printability Through Focus ............................................................................ 55
6.6 Summary ............................................................................................................. 56
7 Compensation Methods for Buried Defects ............................................................... 57
7.1 Compensation with Design Curves ..................................................................... 57
7.1.1 Development of Design Curves ..................................................................... 58
7.1.2 Example of Compensation with Design Curves ............................................ 60
7.2 Compensation by Defect Covering ..................................................................... 61
7.2.1 Example of Compensation by Defect Covering ............................................ 61
7.2.2 Example of Compensation by Defect Covering for a Narrower Defect ........ 63
7.3 Compensation of Buried Pits............................................................................... 64
7.4 Summary ............................................................................................................. 65
8 Conclusion .................................................................................................................. 66
9 References .................................................................................................................. 68
1
1 Introduction 1.1 Motivation
In 1965, Intel founder Gordon Moore observed that the number of components that could
be fabricated into a single integrated circuit for a fixed cost had approximately been doubling
each year, and he predicted this rate of development would continue. It turned out that this
prediction has been pretty accurate, and the continuing exponential rate of growth is referred to
as Moore’s Law. Maintaining this rate of growth has required constant and coordinated hard
work, innovation, and investment. The most critical technology for maintaining this growth is
lithography. Current ArF lithography technology is reaching its limits and the lithography
community is working hard to replace it. The leading contender to replace ArF lithography,
which uses 193nm light, is extreme ultraviolet (EUV) lithography, which uses 13.5nm light.
Unfortunately for the industry EUV lithography, the topic of this dissertation, faces many
challenges. Currently the primary challenge is EUV mask defectivity.
Lithography transfers the mask pattern to the wafer; therefore a fundamental assumption of
lithography is that the pattern on the mask is correct. If this is not the case, and there is a defect
on the mask, it could be transferred to the wafer resulting in a useless integrated circuit. Luckily,
not all mask defects cause unacceptable changes to the wafer pattern. Assuming any
imperfection on the mask will result in a non-working circuit would make mask fabrication
impossible. In reality the size, shape, location and material properties of a defect determine
whether or not the defect will actually be a problem. Understanding EUV mask defects is made
more complex because at the EUV wavelength low material contrast requires additive reflections
from many layers stacked to form a multilayer to achieve adequate mask reflectivity. The local
reflectivity is vulnerable to defects buried in the multilayer. This type of defect, referred to as a
buried defect, is unique to EUV. Studies to understand the interactions of all of a defect’s
characteristics with mask features must be performed well in advance of production equipment
availability. Because production equipment is not available for these studies, simulation is
necessary to produce the needed information during the technology development process.
This dissertation describes the development and application of a new simulator,
RADICAL, which can accurately simulate the interaction of mask features and a buried
multilayer defect three orders of magnitude faster than rigorous methods. RADICAL achieves
this performance by using simulation and modeling methods designed specifically for the
individual EUV mask components to be simulated. One of the components, developed by
Michael Lam in his Ph.D. thesis, is a ray tracing simulator to predict the scattering from a
multilayer with a buried defect. The other is a propagated thin mask model to predict the
diffracted light produced by a plane wave incident on an arbitrary absorber pattern. Predicting
this diffraction required the careful study of rigorous simulation results to correctly model the
effects of the absorber thickness and edges on the magnitude and phase of the diffracted orders.
This modular design allows multilayer simulation outputs to be re-used so that multiple absorber
patterns over the same buried defect geometry can be simulated rapidly.
RADICAL is configured to simulate a wide assortment of EUV defects and mask patterns.
RADICAL determines the nearfields just above a reflective mask for an incident plane wave at
an arbitrary angle of incidence. The diffraction spectrum from these fields can then be input into
2
a commercial aerial image software package. In this dissertation, Panoramic EM-Suite is used to
determine the aerial image at the wafer including illumination, demagnification, defocus and
aberrations.
Currently, EUV mask blank manufacturers are unable to produce a defect free multilayer.
In fact, the industry is struggling to even define which blank imperfections are tolerable, and
which are unacceptable defects. RADICAL has provided important new physical understanding
and quantitative data to assist with these industry decisions. Current EUV exposure tools are
also not yet at production quality. RADICAL simulations have been compared to experimental
results from the Actinic Inspection Tool (AIT) at Lawrence Berkeley National Laboratory
(LBNL) to give insights on both the characteristics of the mask defects and the tool itself.
Another application of RADICAL has been to explore compensation methods for buried defects
in EUV masks. The goal of this compensation work is to adjust the absorber pattern near
unacceptable defects such that the image transferred to the wafer does not show the effects of the
defect.
It is important to note that EUV technology continually evolved as the work for this
dissertation was performed. This has been exciting because there has been a lot of interest from
industry in this work and it has been rewarding to play a role in the technology evolution.
Nonetheless, focusing on the most relevant issues and finding reliable results to compare with
has been a challenge. Initially, there was limited experimental data to compare to, and the data
that was available used tools that were under development themselves. For example, the quality
of the AIT tool changed considerably making each comparison a challenging new calibration.
Also, the multilayer deposition process varied significantly over time and between
manufacturers, making it challenging to interpret and compare to results in the literature. Thus,
the goal of this dissertation was to study the most difficult simulation problems whenever
possible. Even if the geometries studied in this work are not identical to what will be used in
EUV production, the methods developed here will be able to quickly and accurately simulate the
effects of future defects.
1.2 Dissertation Content and Contributions
This dissertation begins by describing the basics of a generic projection lithography
system. Following that are the details of extreme ultraviolet (EUV) lithography with a focus on
what makes the technology so difficult to implement. The final section of Chapter 2 describes
the current state of EUV lithography research tools with special attention given to The Actinic
Inspection Tool (AIT) at Lawrence Berkeley National Lab. AIT inspection images will be
compared to RADICAL simulation results several times in this dissertation.
Chapter 3 summarizes the history of defective EUV mask simulation. Because of the
complex geometry, simulating an EUV mask with a buried defect is computationally intensive
using rigorous simulation methods. There have been many attempts to reduce the computational
requirements of defective EUV mask simulation, but none have been fast and accurate enough
for detailed, reliable studies of EUV defect printability and compensation. After this historical
context is provided, the new simulator RADICAL is introduced and its design, computational
requirements and accuracy are described in detail. RADICAL is by far the most important
contribution of this dissertation. It is 1,000 times faster than FDTD for simulating defects within
a line space pattern and uses two orders of magnitude less memory. In the process of studying
the sources of error from the various components of RADICAL, it was realized that if a simple
top surface model, called the single surface approximation (SSA) were improved to model the
3
penetration into the multilayer and variation with incident angles, it could be used in place of the
ray tracing simulator in RADICAL for shorter defects with minimal multilayer smoothing in
patterns with a half-pitch above 22nm. If this advanced SSA model is integrated into RADICAL
the runtime is increased by another two orders of magnitude. The sections of this dissertation
after Chapter 3 present the first thorough study of buried defects for the 22nm and 16nm
technology nodes. Without the speed increase provided by RADICAL the extensive quantitative
studies found throughout this dissertation would not be possible.
Chapter 4 is devoted solely to the study of isolated buried defects. Defects in EUV
lithography are problematic when they are located near features on the mask and cause these
features to print incorrectly. But, Chapter 4 shows that studying isolated defects gives valuable
insight into the printability of buried defects. The inversion of the aerial image of an isolated
buried bump defect through focus, from a dark spot for negative focus to a bright spot for
positive focus is a defect characteristic that will be important in every portion of this dissertation.
Chapter 4 also includes comparisons of RADICAL and experimental AIT images. These
confirm the accuracy of the multilayer simulation method in RADICAL. The main contribution
of this chapter is that isolated defects do not behave like conventional thin mask defects with a
constant phase or magnitude. This is important because it shows that although the reflected
phase difference due to the defect is the dominant characteristic, referring to defects by a single
lateral size or phase value will not adequately describe them.
Chapter 5 shows an interesting application of RADICAL for determining aberrations in the
AIT. The center intensity of a buried defect image through focus turns out to be sensitive to
aberrations. Different aberrations, such as astigmatism, spherical and coma, affect the through
focus intensity in different ways. This is exploited in a new method for determining large
aberrations using through focus images of an isolated EUV defect. This method is interesting,
and demonstrates the effects of aberrations on buried defects. But, it is not as important as other
parts of this dissertation because it will likely not be as precise as other methods for the small
aberrations expected in future EUV tools.
Chapter 6 directly investigates the printability of buried defects near features in various
future production scenarios for 22nm and 16nm dense lines. The printability of buried defects
depends on many geometric characteristics, such as the size, shape and position relative to the
features. For both 22nm and 16nm dense lines some defects shorter than 1nm tall on the
multilayer surface can cause greater than 10% critical dimension (CD) change in focus. Out of
focus, the defect tolerance is even worse. Comparisons to AIT simulations of defects near
features will confirm many of the trends observed in the simulation results. This chapter
provides the first thorough look at defect printability at the 22nm and 16nm nodes. It shows the
strong dependence of CD change on the defect position relative to the absorber lines, including
the fact that a defect only a couple nanometers tall located under an absorber line can cause a
greater than 10% CD change.
While Chapter 6 shows that multilayer defects in EUV masks are a serious problem,
Chapter 7 proposes a solution. The topic of Chapter 7 is defect compensation by adding or
subtracting absorber near the defect. It turns out that by using pre-calculated design curves,
several possible compensation geometries can be prescribed based only on the CD change
caused by the buried defect. If this compensation is applied, the effect of the defect at best focus
is nearly eliminated. But, through focus the defect still causes a CD change. A method is
proposed to reduce this through focus CD change as well. These methods to compensate for
buried defects may relax the mask blank defectivity requirements. If the necessary metrology
4
and repair tools are available, the methods in this chapter show how simple absorber adjustments
can compensate for a buried defect.
5
2 EUV Lithography 2.1 Lithography for Semiconductor Manufacturing
Optical projection lithography is currently used by all leading semiconductor
manufacturers to faithfully and economically transfer a designed circuit pattern to a silicon
wafer. Projection lithography has many advantages over competing technologies. Unlike
contact printing or nanoimprint lithography, there is no physical contact between the mask and
wafer, which greatly reduces the defects added during printing. Unlike in proximity printing, the
optics between the mask and wafer in projection lithography are designed to collect the
diffracted light and focus it onto the wafer, and resolution is not hurt by Fresnel diffraction.
Projection printing is parallel, meaning relatively large sections of the pattern are printed
simultaneously, which sets it apart from electron beam lithography. Finally, in projection
printing the source and mask can be designed to work together with the projection optics to
improve the resolution to well below the wavelength of the light used.
A generic projection lithography system, in this case with a dipole illumination, is shown
in Figure 2-1. There are three major components of this system: the illumination, which is made
up of the source and condenser optics, the photomask, and the projection optics. The
illumination projects light through the mask and projection optics to form an aerial image on the
wafer. The aerial image is the intensity distribution at the wafer. The aerial image is recorded
by the photoresist, allowing the mask pattern to be transferred to the wafer. The time average
work done on the resist, which is what the resist records, is proportional to this intensity, which
is given by the electric field squared, not the electric field. For this dissertation, an aerial image
will be considered the final output of this system.
Figure 2-1. Schematic of a generic lithography system. A dipole source is shown, with the light from two of the
many radiators in each pole traced to the mask.
Source (Dipole)
Condenser Optics
Photomask
Projection Optics
Wafer
6
2.1.1 Illumination
Sources and condenser optics for lithography scanners are very complex. But, for this
work the illumination system will be treated as a black box with an output defined by its source
distribution. When the terms “source distribution” or “illumination” are used in this dissertation
they refer to a distribution of independent radiators in the source plane. The purpose of the
condenser optics is to produce a uniform plane wave on the mask from each of these radiators in
the source plane. Figure 2-1 shows a dipole with two of the radiators highlighted. In reality,
every radiator in the source is producing a plane wave incident on the mask at a slightly different
angle. Each of these radiators is mutually incoherent, meaning that the phase of the light emitted
from one is not related to the phase of any other. The light from each radiator travels through the
entire optical system and forms an image on the wafer. Because the phase of each radiator is
independent of the other radiators, the sum of the intensities of the resulting images is what
interacts with the resist on the wafer. Incoherent illumination is used because it results in a
smoother image, without ringing or speckle, than a coherent source would.
There are several different illuminations that are commonly used. Three examples are
shown in Figure 2-2. Three examples of source distributions used in lithography. Figure 2-2.
The white areas represent where light is output by the source. The center represents light which
ends up being normally incident on the mask. Away from the center, the larger the radius the
higher the incident angle of the wave. Light incident at higher angles is capable of printing
smaller features than normally incident light, with some limitations. For example, a dipole
illumination can print line space patterns very well, but only in one direction.
Figure 2-2. Three examples of source distributions used in lithography.
The simplest metric to describe an illumination is the partial coherence factor, sigma (σ).
This is defined as the ratio of the sine of the maximum angle of light incident on the photomask
from the source to the sine of the maximum angle of incidence collected by the projection optics.
The sin of the maximum angle collected by the projection optics is defined as the numerical
aperture, NA, which will be described later in this chapter.
If an illumination has a high sigma value, then light is incident on the photomask at high
angles. The partial coherence factor is normally below one. A partial coherence value of zero
means that the mask is illuminated by a normally incident plane wave.
Top Hat Annular Dipole
7
2.1.2 Photomask
The photomask, also referred to as a reticle, is the master copy of the circuit layout which
is transferred to every circuit printed. There are two basic types of photomasks: binary and
phase-shifting. A binary mask simply blocks the light in some areas of the mask and not in
others. For example, if a mask is to be used to print a metal wiring pattern on the wafer, a
location on the mask where the light is blocked could correspond to a location on the wafer
where there is no metal. Conversely, a point where the light is not blocked will correspond to a
location on the mask where there is metal.
The second basic type of mask is a phase shift mask. Unlike a binary masks, which blocks
light, a phase shift mask changes the phase of incident light as it is transmitted through the mask.
This allows for higher diffraction efficiency into desired orders, because the light is not blocked
as in a binary mask.
2.1.3 Projection Optics
The projection optics are what deliver the image from the mask on to the wafer. In most
lithography systems the projection optics reduce the size of the mask image by a factor of four
when printed on the wafer. There are two characteristics of the projection optics that will be
important for this dissertation: numerical aperture (NA) and the level of aberrations.
The numerical aperture is defined as:
Equation 2-1: �� � � sin ��
where n is the refractive index of the material the image is projected into and θmax is the
maximum angle output by the projection optics. The NA of the projection optics is directly
related to the resolution (R) of the lithography system and the depth of focus (DOF)
Equation 2-2: � � ��
��
where λ is the wavelength of the light and k1, the “technology factor”, accounts for all
factors other than wavelength and NA that affect resolution. For example, using the annular or
dipole illuminations shown in Figure 2-2 can reduce the k1 factor.
The wavelength and NA of the system also determine the depth of focus (DOF) of a
lithography process. Like resolution, depth of focus depends on more than just the wavelength
and NA, so it is defined as shown in Equation 2-3.
Equation 2-3: ��� � � ��
�����
where k2 is another factor, like k1, that includes many additional parameters beyond
wavelength and NA. A typical k2 value is 0.5. With k2 = 0.5, the depth of focus becomes one
Rayleigh Unit, defined by Equation 2-4.
Equation 2-4: ������� "��# ��
�
�
�����
The depth of focus is the amount the plane the image is being projected onto can move
nearer or farther away from the projection optics without producing an unacceptable image.
Normally this plane the image is being projected onto is in the resist on top surface of the wafer.
Having a larger DOF is advantageous because it means less control of the wafer position and
surface uniformity is required in the scanner.
Two characteristics determined by the projection optics are critically important to
lithography. One is resolution through a usable DOF. The other is the level of aberrations.
High quality projection optics have very low aberrations. Aberrations are imperfections in the
8
optics which cause the printed image to deviate from the desired image. A photomask diffracts
the light incident onto it over a large angular spectrum. Ideally, the optical path length would be
the same for each of these angles. Aberrations are non-idealities that cause different path lengths
for each angle. This causes the phases of the waves that form the image being incorrect and
causes errors in the wafer image. Aberrations are a serious problem for current EUV actinic
inspection tools.
2.2 Lithography with Extreme Ultraviolet Light
The primary goal of the lithography industry is to follow Moore’s Law by continuing to
shrink the sizes of features printed on integrated circuits. Shrinking feature sizes requires
improving, or decreasing, the possible resolution. As Equation 2-2 shows, to improve the
resolution of a lithography system the numerical aperture must be increased or the wavelength or
k1 must be decreased. Current and proposed EUV systems have a lower NA and simpler
illumination than current deep ultraviolet (DUV) tools. But, they are still able to achieve better
resolution because the wavelength is decreased significantly from 193nm to 13.5nm. This lower
wavelength, along with lower NA, gives an improvement in resolution and depth of focus.
The most advanced currently available 193nm scanners have a numerical aperture of 1.35.
Which corresponds to a Rayleigh unit, or approximate DOF from Equation 2-4, of 53nm.
Production EUV lithography is expected to be introduced with an NA of 0.32. This corresponds
to a Rayleigh unit of 66nm. So there is a small improvement DOF. The required k1 from
Equation 2-2 to print 22nm lines, for example, increases from 0.15 for 193nm lithography to
0.52 for EUV lithography. To print with a k1 as low as 0.15 will most likely require double
patterning or simultaneous optimization of the mask and source, which will be expensive and
difficult [1, 2]. Comparatively, printing with a k1 of 0.52 is fairly simple. Only mild optical
proximity correction (OPC) is required [3]. Another advantage of a lower NA value is that
polarization effects, on the wafer and mask, are greatly reduced because the variation in the
angles of incidence of the light on the wafer and mask is much lower.
The description in section 2.1of a lithography system was generic and could apply to any
form of projection lithography. To understand this dissertation it is important to understand the
details of EUV lithography.
2.2.1 Illumination for EUV Lithography
The wavelength of light used for EUV lithography is 13.5nm. To generate light at this
wavelength, production tools will use tin plasma. There are two competing technologies to
generate this plasma: laser produced plasma (LPP) and discharge produced plasma (DPP). The
differences between these two methods are beyond the scope of this dissertation. It is important
to note that compared to the ArF sources used for DUV lithography, EUV sources generate a
relatively large wavelength spectrum. Therefore, the mirrors between the source and the wafer
are the main determinant of the bandwidth which contributes to the final wafer image.
EUV illumination systems are designed to be the same as the illumination systems
described in 2.1.1. Current tools, designed to print features in the 30nm range use mainly tophat
illuminations, because off-axis illumination are not required for the desired resolution. When
EUV goes into production, off-axis illuminations like those in Figure 2-2 will likely be used.
9
2.2.2 Photomask for EUV Lithography
All EUV lithography masks discussed in the dissertation, and most EUV masks used in
practice, are binary masks. The major difference between a conventional lithography mask and
an EUV mask is a conventional mask modulates transmitted light, and an EUV mask modulates
reflected light. A reflective mask is necessary because any mechanically rigid transmitting mask
would absorb too much EUV light to be usable. This absorption is a problem at EUV
wavelengths and not at DUV wavelengths because at EUV no materials are available that would
not unacceptably lossy. All materials have a refractive index around one and a significant
imaginary component. This imaginary component causes absorption of propagating light. The
structure of the mask is shown in Figure 2-3.
Figure 2-3. Sideview cutline of an EUV mask (left) and the resulting aerial image (right).
An EUV masks has two main components which interact with the incident light. The first
is the multilayer, which reflects the light incident on it. An EUV mask multilayer is usually
composed of many layers of Silicon (Si) and Molybdenum (Mo) stacked alternately on top of
each other. A pair Si-Mo layers is referred to as a bilayer. These bilayers are required because
at a wavelength of 13.5nm, the material differences between Si and Mo is so small that a single
layer would reflect a very small percentage of the energy. The geometry of the multilayer is
chosen so that the small amount of light reflected from each bilayer adds in-phase with the light
reflected from all of the other bilayers. Ideally for a multilayer composed of 40 bilayers, which
is a common design, about 75% of the incident power is reflected. In practice, reflectance values
are around 70% [4].
The second component is the absorber, which defines the pattern to be transferred to the
mask. The absorber is several wavelengths tall and can be made out of several different
materials. In this dissertation, the absorber is assumed to be made out of Tantalum Nitride (TaN)
with an anti-reflective coating on top. The anti-reflective coating is necessary for inspections of
the mask using tools designed for DUV masks. The absorber works very simply: where there is
absorber most of the light is absorber (i.e. not reflected) and where there is no absorber the light
is reflected.
Refractive Indices of Simulation Domain
0 100 200 300 400 500 600 700
0
50
100
150
200
250
300
350
400
450
10
2.2.3 Projection Optics for EUV Lithography
Like the mask, optics for EUV lithography must be reflective. There are advantages and
disadvantages to reflective optics. One major advantage is the lack of chromatic aberration. In
refractive optics, light at different wavelengths are focused to different positions because of the
slightly different refractive index values for different wavelengths. This aberration is not present
in reflected optics because light is always reflected at the angle it is incident on a surface,
independent of wavelength. One problem with reflective optics at EUV is that each mirror absorbs a non-trivial amount
of energy. Because sources can only produce a limited amount of power, the number of mirrors
that can be used for the optics is limited. Also, the decreased wavelength means that much
smaller physical non-idealities, such as mirror non-uniformity, will cause unacceptable
aberrations. But, current projection optics are meeting the aberration requirements and the
quality of the projection optics is not currently an issue for EUV lithography [4, 5].
Another issue with reflective optics is that finding a path for the light around all the
necessary mirrors is difficult. For NA above 0.4, a 6-mirror system will not be possible [4].
Either more mirrors will be needed, or the design will have to be fundamentally changed with the
addition of a central obscuration. A central obscuration means that the angles near the 0th
order
reflected off the mask are lost and do not contribute to the final wafer image. An example of a
real system with a central obscuration is the Berkeley Micro-exposure Tool, which is described
in Section 2.4.2.
2.3 Current issues in EUV Lithography Development
The advancement of EUV lithography has been slow and the expected insertion time has
continued to move farther and farther into the future. The primary hindrance to EUV lithography
is economics. EUV lithography systems could be installed in a semiconductor fabrication
facility in their current state would make some working chips. But, the cost of the system would
be much, much higher than the technologies currently being use. There are three main
components of the system that have kept EUV lithography from being implemented: the source,
photoresist and mask. The problems with each of these are described below. Since
understanding EUV mask defects is critical to understand this dissertation, they will be covered
in more detail here than the source and photoresist.
2.3.1 Source
The main problem with the EUV source is power. Source power is directly related to
throughput. If the power is high, each wafer must be exposed for a short amount of time. If the
power is decreased, each wafer must be exposed for longer, and the throughput is decreased.
The useable EUV power level needed for high volume manufacturing is several hundred watts.
But, for the past several years the highest sustained power produced by any the source has been
10W or less. Recently, however, source powers have increased significantly and source
manufacturers are confident that the source will not prevent the implementation of EUV
lithography. [6]
2.3.2 Photoresist
Photoresist, usually referred to simply as resist, is the photosensitive material which is
deposited onto silicon wafers before each lithography step. The photoresist’s chemical
11
properties are changed when it is exposed to light, which allow it to temporarily record the
pattern on the wafer in between process steps. There are three figures of merit for resist that must
be balanced and optimized: sensitivity, resolution and line edge roughness (LER). Sensitivity,
measured in joules/cm2, determines how much energy is required to transfer the pattern to the
resist. Resolution, measured in nm, is the smallest pattern a resist can print. Line edge
roughness, also measured in nm, is the high frequency (compared to the feature sizes) variation
in the position of the edge of a resist line. As with many things in engineering, it is not difficult
to improve two of these figures of merit at the expense of the third. But, improving all three
simultaneously is difficult. Currently, EUV resist performance appears to be on track to meet
overall EUV lithography targets, though there are ominous signs improvement is slowing [7].
2.3.3 Mask
As explained above, a reflective EUV mask is completely different than a conventional
transmission mask. This presents two related problems which are a significant hindrance to the
implementation of EUV lithography: mask defectively and mask inspection. Currently, EUV
mask blank suppliers are unable to produce defect free masks [8]. A mask blank is an EUV
mask without a pattern printed on it. The defects in these blanks occur in the multilayer.
Sometimes they are on the substrate, perhaps due to a scratch which occurs during substrate
polishing. Other times the defects are due to particles which are unintentionally deposited along
with the multilayer deposition process. These sorts of defects are often referred to as buried
defects. They are also sometimes referred to as phase defects, because their primary, but not
only, effect is on the phase of the reflected field. Two examples of buried defects are shown in
Figure 2-4.
Figure 2-4. Two computer generated cross-section refractive index map examples of an EUV mask blank with a
buried defect. The left image is of a substrate pit or scratch. The right image is of a particle which was deposited
during the multilayer deposition.
Current reported blank defect densities in EUV masks are on the order of one defect per
centimeter squared [9, 10, 11], which is unacceptably high. However the second issue, mask
inspection, actually calls into question the legitimacy of the defect densities above. The actual
density of printable defects is likely much higher. Mask inspection tools are used to detect
defects on lithography masks. These tools are very expensive to design and manufacture. For
X-axis (nm)
Y-A
xis
(nm
)
0 50 100 150 200 250 300 350 400 450 500
0
50
100
150
200
250
300
350
400
450
X-axis (nm)
Y-A
xis
(nm
)
0 50 100 150 200 250 300 350 400 450 500
0
50
100
150
200
250
300
350
400
4500.8
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
12
this reason, there are no mask inspection tools currently available which are able to detect detects
as small as the smallest printable defects for EUV lithography. This means that currently
reported defect densities are lower than the actual printable defect densities on the mask. The
current inspection tools miss defects that will negatively affect yield in EUV lithography.
2.4 Existing EUV Lithography Tools
EUV lithography tools suitable for high volume manufacturing are not currently available.
But, there are many EUV research tools around the world which are currently being used to help
develop the technologies, such as masks and resist, necessary for EUV to be implemented.
These tools, especially the Actinic Inspection Tool, have been essential for providing
experimental verification of this dissertation. Each of these tools falls into one of three
categories: full field scanners, micro-exposure tools and actinic inspection tools.
2.4.1 Full Field Scanners
Full field scanners are the types of tools which will be used to manufacture integrated
circuits if EUV lithography is ever adopted. Currently the leading scanner manufacturer, ASML,
has two alpha tools in the field [12]. These tools are being used for all sorts of EUV research
such as mask defect printability [9,10] and for demonstrations of device integration with EUV
[13,14].
2.4.2 Micro Exposure Tools
Micro exposure tools, like scanners, are used to expose resist on wafers [15,16]. But,
unlike scanners and as the name implies, they only expose a very small area of the wafer.
Currently, the primary use for micro exposure tools is resist evaluation.
2.4.3 Actinic Inspection Tools
There are several types of actinic inspection tools. What they all have in common is they
expose an EUV mask with actinic light. In this case, actinic means the wavelength of the light is
approximately the wavelength that is used in an EUV scanner. One kind actinic inspection tools
are dark field [17]. These tools illuminate an EUV mask with an on-axis plane wave and form
an image on a detector with the light scattered off axis. The on-axis light does not contribute to
the image. Bright field actinic inspection tools are also used for mask research [18, 19, 20].
These tools illuminate a mask with a plane wave or more advanced illumination and collect all of
the light scattered from the mask within a pupil defined by the NA of the tool to form the aerial
image on a detector. These sorts of tools are valuable for the research in this dissertation,
because the fields can be compared directly to aerial images predicted by simulation without
needing to model complicated resist behavior. For this reason, significant comparisons were
made between the Actinic Inspection Tool (AIT) at Lawrence Berkeley National Laboratory
(LBNL) and RADICAL, the simulator described in this dissertation.
2.4.3.1 AIT at LBNL One of the leading actinic inspection tools in the world is at Lawrence Berkeley National
Laboratory (LBNL). The tool is generally referred to as The Actinic Inspection Tool or AIT.
The AIT is a Fresnel zoneplate microscope designed to emulate the optics of an EUV scanner
[18]. There are five different zoneplates built into the tool. Any one of the zoneplates can be
13
used to emulate a certain imaging condition. The image formed by the zoneplate is projected
onto a charge-coupled device (CCD) to record the image. The source for the AIT is a bending
magnet on the Advanced Light Source (ALS) at LBNL. The light received by the AIT has a
very narrow bandwidth, but the wavelength itself can be adjusted. This is a valuable capability
and is currently used as a robust way to adjust the focus position of the AIT image [21].
There are several important differences between the AIT and a production EUV scanner.
The one with the largest effect on the final image is partial coherence. A production EUV
scanner will have a partial coherence value (σ) of anywhere from 0.5 to nearly 1.0 and may use
advanced illuminations such as annular or dipole. But, the AIT illumination is fixed and much
more coherent than a production scanner. In this work, it is modeled as a tophat illumination
with σ=0.1. Another different between the AIT and a production tool is the bandwidth of the
light. A production tool will have a much wider bandwidth than the AIT. The final major
difference is aberrations. The fundamentals of zoneplate imaging allow at best only a single
point to be without aberrations, away from this point aberrations increase.
2.4.4 Programmed Defect Mask (PDM)
As explained above, buried defects in EUV masks occur naturally and more often than is
acceptable. But, studying only these randomly occurring defects makes systematic experimental
studies of defect printability impossible. Therefore, program defect masks (PDM) are created to
allow systematic experimental studies of buried defect printability [22,23,24]. EUV masks with
programmed buried defects are created be depositing or removing some material from the mask
substrate before multilayer deposition, creating defects. Then the multilayer is deposited on top
of these defects and the resulting size of the defects on the surface is measured. Once the defect
sizes are measured, the desired absorber pattern is deposited on top of the multilayer.
Intel fabricated a program defect mask to study EUV defects [22]. All AIT inspection
results analyzed in this dissertation were done on this programmed defect mask in which 48nm
high posts with a square base on a substrate were over-coated with a multilayer. The width of
the posts was varied. It turns out that the smoothing process used for the multilayer deposition
produced defects which all have between a 50 and 60 nm full width half max (FWHM) diameter
and heights ranging up to 8nm. The substrate and surface sizes of all the defects on the mask is
given in Table 2-1. This mask has only substrate bumps, not pits. Therefore, bumps will be the
focus of comparisons between experiments and simulations in this work.
Table 2-1. Defect sizes for the programmed defect mask. Buried defects are boxes on the substrate with a square
base and constant height. The surface defects are the result of the multilayer smoothing and are roughly Gaussian
shaped.
Buried Width Buried Height Surface FWHM Surface Height
100 48 60 8
95 48 59 7
90 48 58 6.2
85 48 56 5.3
80 48 55 4.4
75 48 54 3.5
70 48 53 2.7
65 48 52 2
60 48 51 1.7
55 48 50 0.8
50 48 49 0.4
14
2.5 Summary
Lithography with extreme ultraviolet 13.5nm light is challenging for several reasons. The
13.5nm wavelength requires significant new development of source, photomask, optics and
photoresist technologies. But, the advantage of much higher resolution with improved depth of
focus keeps the industry working hard to implement EUV lithography successfully for integrated
circuit manufacturing. The simulation tool development and defect characterization work that
will follow in this dissertation has and will continue to play a guiding role in this development
work.
15
3 Fast Simulation Method
for EUV Masks with
Buried Defects 3.1 History of EUV Mask Simulation
The history of defective EUV mask simulation is extensive and has included several
methods to accelerate the simulation process. EUV masks are different than conventional
lithography masks. The main difference is that EUV masks are reflective. A multilayer mirror is
used to reflect incident light and this mirror is susceptible to buried defects. Conventional
lithography simulation methods are not designed to simulate the complex defective multilayer
structures, so a lot of work has been done to develop simulation methods specifically for EUV
masks with buried defects.
One type of simulator used extensively in conventional lithography that can be applied
directly to EUV defective mask simulation is a rigorous simulator. Rigorous simulation methods
have been used for EUV masks with buried defects. A common rigorous and scalable method is
the finite difference time domain (FDTD) [25]. Unfortunately, for EUV masks FDTD is very
computationally intensive and has questionable accuracy for off-axis angles of incidence [26].
Modal methods, like rigorous coupled wave analysis and the waveguide method, have been
shown to perform faster than FDTD. But, unlike FDTD, the runtime of modal methods depends
greatly on how well the geometry can be approximated by rectangles, reducing the speed
improvements over FDTD for defective EUV multilayer simulations [ 27 , 28 ]. Another
commonly used rigorous method shown to be faster than FDTD is the finite element method
(FEM) [29]. FEM models the geometry with a non-uniform triangular mesh so it can simulate
geometry irregularities, like a buried defect, with less affect on runtime than the waveguide
method [30].
Dramatic speed and accuracy improvements over rigorous methods have been shown by
linking analytic multilayer models with FDTD [31,32], but these methods do not account for a
defective multilayer. Many approximate methods have been developed that do account for
buried defects. In [33], Bollepalli simulated an EUV mask with a buried defect using code
optimized for proximity X-ray lithography. In [34] Evanschitzky used the approximation in
[35], that a defective multilayer is made up of subregions of defect free multilayers of different
heights lined up next to each other, along with FDTD to simulate a full EUV mask with a buried
defect.
The fastest simulation method for defective multilayer simulations is the single surface
approximation (SSA). It was presented by Gullikson, along with an algebraic printability model
in [36]. But, SSA used in conjunction with the algebraic imaging model was shown to over
predict the effect of the defect on the aerial image for defects with a surface width larger than
30nm [37]. This was because the major assumption of the imaging model is that the scattering
from the defect filled the pupil uniformly. But, for defects larger than 30nm wide this is not true,
16
the pupil is not filled uniformly. For these larger defects, the light is concentrated at the center of
the pupil. Therefore, assuming that the intensity value of the zero-order is uniform across the
pupil, as the algebraic model in [36] does, predicts an effect much larger than is actually
produced by the defect. If the near fields predicted with SSA and third party imaging software,
such as SPLAT , Prolith or Panoramic EM-Suite, are used in place of the algebraic model to
calculate the aerial image, SSA under predicts the impact of most buried defects and does not
model defect feature interaction [26]. In this dissertation a new method for tuning and
generalizing SSA to arbitrary angles of incidence, referred to as Advanced SSA, is introduced.
This method is shown to be sufficiently accurate in many EUV mask geometries.
A ray tracing method for multilayer blank simulation as accurate as FDTD, yet three orders
of magnitude faster than FDTD was developed and evaluated by Lam in [38]. A generalization
to a propagated thin mask model for EUV mask absorber features was proposed in [39]. These
methods are a major part of this work and will be expanded upon below.
The first goal of this dissertation is to develop fast, accurate and flexible methods for
simulating the challenging interactions between EUV blank defects near absorber features. Each
of the simulation methods discussed above was a breakthrough for EUV mask simulation. But,
none of them were simultaneously fast, accurate and flexible. For example, FDTD is very
accurate and is flexible enough to simulate any geometry, but it is slow. Ray tracing, on the
other hand is very fast and accurate for multilayer simulations, but its flexibility is limited by an
inability to simulate defects near features. For this reason, the new simulator RADICAL was
created.
Unless specifically noted otherwise, all simulation methods discussed, like RADICAL,
SSA and FDTD, are used to predict the electric field reflected from the EUV mask. To produce
the aerial image at the wafer a separate aerial image simulator is necessary. For this work,
Panoramic EM-Suite is used.
3.2 RADICAL
RADICAL, which stands for rapid absorber defect interaction computation for advanced
lithography, is a simulation program designed specifically for fast, accurate and flexible
simulations of EUV masks with buried defects. It was presented in [40] and can simulate EUV
masks with buried defects and absorber features two orders of magnitude faster than FDTD using
two orders of magnitude less memory. This speedup is accomplished by simulating the absorber
features and defective multilayer separately using simulation methods optimized for each. The
simulator flow is shown in Figure 3-1. The absorber layout simulator runs without regard for the
multilayer geometry and the multilayer simulator runs without regard for the absorber layout.
This modularity makes the fast and accurate simulation of the entire mask possible. Both
components of the simulator, the thin mask absorber model and multilayer simulator take an
arbitrary plane wave input and output a complex electric field. A Fourier transform is used to
convert the electric fields output by one simulator into plane waves to be input into the next.
Multiple reflections between the absorber and the multilayer are not considered in RADICAL to
save runtime. This approximation is valid because the reflection of light coming from the
multilayer off the bottom of the absorber pattern is very small and does not contribute a
meaningful amount to the reflected field.
17
Figure 3-1. Graphical summary of the RADICAL simulation flow
RADICAL has been implemented with two different methods for multilayer simulation.
The first implementation, referred to as “standard RADICAL”, uses the ray tracing multilayer
simulator presented in [38]. RADICAL has also been implemented using the advanced single
surface approximation to simulate the multilayer. This is referred to as RADICAL with SSA.
3.2.1 Multilayer Simulation
The modularity of RADICAL makes it trivial to integrate different models into the
simulation flow. Two different techniques are used to simulate the multilayer. For taller defects
with very non-uniform layers below the surface ray tracing is used. The shorter defects with
relatively uniform layers below the surface the single surface approximation can be used. The
details of each method are explained below.
3.2.1.1 Ray Tracing The ray tracing multilayer simulator was presented in [38]. It is capable of simulating a
defective EUV mask blank which is a multilayer with no absorber features. The simulator uses
ray tracing to simulate EUV light propagating into the multilayer. The key observation made by
Lam while creating this simulator was that each silicon molybdenum pair could be treated as a
single layer, because the two layers within the bilayer are nearly parallel. Therefore a layer that
is tilted, due to a buried defect, causes the transmitted ray to be laterally shifted slightly. This
accurately accounts for the disrupted geometry of the multilayer due to the defect. The
propagation out of the multilayer is done differently. For this step, the multilayer is assumed to
be perfect. The fields at each layer, predicted by ray tracing, are converted to their Fourier
components and propagated out analytically. This step, which is a perturbational approach,
accounts for the resonances in the multilayer by summing the infinite series of reflections in
closed form.
Reflection from
multilayer
Transmission through
absorber features
Ray
Tracing
Thin Mask
Model
Plane Wave
Near Field
Incident
Wave FT{ }
Near
FieldFT{ }Absorber Layout
SimulatorFinal Result
Multilayer
Simulator
Absorber Layout
Simulator
Near
Field
Absorber Pattern
Specifications
Defect and Multilayer
Specifications
Fourier
Transform
Reference
Plane
SSAor
18
These approximations make the ray tracing simulator much faster than rigorous methods
without loss of accuracy for most substrate defects [26]. Though, if layers within the multilayer
become overly distorted, which could happen for large substrate defects, the perturbational
approximation for propagation out is not valid and rigorous simulations may be necessary.
3.2.1.2 Single Surface Approximation and Its Generalization The single surface approximation (SSA) is a method to predict the electric field reflected
from a defective EUV mask blank. It was introduced in [36]. The single surface approximation
assumes that a buried defect is purely a phase defect, and the change in phase is determined by
the round trip path difference due to the perturbation of the surface by the buried defect. It
ignores the effects of all of the layers below the top surface of the multilayer. It was shown in
[41] that using a slightly larger surface defect height than is actually on the mask, to account for
the penetration of the light into the multilayer, improves the accuracy of SSA for modeling
isolated defects illuminated with normal incidence plane waves.
To use SSA in RADICAL, it must be able to simulate the reflected field for plane waves
incident at arbitrary angles. Changing the angle of incidence of a plane wave on an EUV
multilayer primarily changes two attributes of the reflected field. It adds a linear progression in
phase across the field and scales the magnitude and phase of the entire field uniformly. This
complex scaling is due to the variation in reflection from the EUV blank as a function of incident
angle and can be predicted accurately analytically.
The advanced single surface approximation is a four step algorithm:
Perform a tuned SSA simulation as described in Section 4.2.4.
Take the Fourier transform and shift the result in k-space to account for off axis angles of
incidence
Scale the complex field by the magnitude and phase of the incident wave.
Scale the complex field by the magnitude and phase of the analytic multilayer reflection
coefficient.
Since the standard SSA method described in [36] does not consider angle of incidence,
step 1 above must only be done once for each RADICAL run. The results can be re-used for all
angles of incidence. Steps 2-4, however, must be repeated for each angle of incidence.
In [41] it was shown that algebraic methods, like advanced SSA, are only valid for defects
that are up to 4.5nm tall on the surface of the multilayer. However, due to substrate and
multilayer smoothing, even substrate defects as large as 50nm tall will produce defects less than
4.5nm tall on the multilayer surface. The assumption that buried defects are purely phase defects
isn’t totally accurate and these inaccuracies are more apparent for out of focus simulations. This
will be discussed more in section 3.2.3 on accuracy.
3.2.2 Absorber Simulation
The absorber, used to define the pattern printed by the EUV mask, is many wavelengths
tall and many wavelengths wide. Because of its large size, relative to the wavelength of light,
and because the differences in the optical parameters of the materials in an EUV mask are small,
a simple model can be used to predict the field transmitted through an absorber pattern. The
development of the propagated thin mask model was presented in [39]. It is assumed that this
calibration is done before a RADICAL simulation, though the 2D FDTD simulation required for
the calibration can be done in less than a minute on a standard laptop computer. Using this fast
and simple 2D calibration allows any EUV absorber stack to be simulated in 3D in RADICAL.
19
Creating the transmitted field from the absorber pattern is a three step process. The first
step is to follow the basic thin mask model and set the field magnitude and phase to the analytic
value below the absorber and to one in the spaces where there is not absorber. The analytic
absorber transmission calculation assumes an infinitely wide absorber, which means edge effects
are ignored. The second step is to add point sources to the edges of the absorber. The Fourier
spectrum produced by an FDTD simulation of an absorber edge shows a constant background of
even orders. The magnitude and phase of the point source is chosen to match the magnitude and
phase of these orders. The final step in the model is to propagate the field with the point sources
approximately half the height of the absorber feature. In [39], for a 70nm absorber, the field was
propagated 40nm. This final step adjusts the phase of the orders in the Fourier transform of the
model to match the FDTD simulation and the exact distance of propagation depends on the
material properties and height of the absorber.
This model has been shown to be accurate for predicting transmission through features as
small as 60nm mask scale, but the accuracy decreases for angles of incidence well above 6° [39].
Therefore this model is accurate for EUV lithography at the 22nm and 16nm nodes. But, it will
need to be re-evaluated and possibly expanded if higher NA EUV systems with nominal mask
incident angles above 6° need to be simulated.
3.2.3 Accuracy of RADICAL
The finite difference time domain will be used for comparisons to determine the accuracy
of RADICAL in two scenarios. The first scenario models a pre-production scanner printing
32nm lines. The other models a production scanner printing 22nm lines. Figure 3-2 shows the
CD change predicted by TEMPEST, an FDTD simulator, standard RADICAL and RADICAL
with SSA for three focus values for the first scenario. RADICAL with SSA uses the tuned SSA
model described in section 4.2.4. The mask pattern is the same 128nm dense line pattern with a
buried defect described in the computational requirements section above. The near fields for
each curve were calculated using each of the three methods. The aerial image from that near
field was calculated using Panoramic EM-Suite for NA=0.25, σ=0.75 and 4x demagnification. A
constant threshold was chosen such that each method has 32nm lines (wafer scale) in the defect
free case.
The main conclusion from this plot is that both versions of RADICAL match FDTD fairly
well for this scenario. For the focus = 0 case, the differences between all three simulators are
0 1 2 3 4
0
5
10
15
20
25
Focus: -108nm
Surface Defect Height (nm)
CD
Change (nm
)
0 1 2 3 4
0
5
10
15
20
25
Focus: 0nm
Surface Defect Height (nm)
CD
Change (nm
)
TEMPEST
RADICAL
RADICAL SSA
0 1 2 3 4
0
5
10
15
20
25
Focus: 108nm
Surface Defect Height (nm)
CD
Change (nm
)
Figure 3-2. CD change predicted by TEMPEST, standard RADICAL and RADICAL with SSA for three focus
values. NA=0.25, σ=0.75,wafer CD=32nm .
20
always less than 1nm, an excellent match. For the out of focus cases, there is a maximum error
of 3.9nm. The assumed application for RADICAL is for many fast simulations of small defects
within a compensation algorithm or to determine maximum allowable defect specifications.
Because the larger defects, where there is some error in RADICAL, will likely cause changes
well above what is acceptable, detailed simulations of their printability will not be necessary. If
very accurate simulations of relatively large defects are required, rigorous simulations will be
necessary.
The error increases as defect size increases. This is expected for both standard RADICAL
and RADICAL with SSA. The ray tracing multilayer simulator assumes that the multilayer is
not too distorted by the defect, it is a perturbational method. As defect size increases, the use of
a defect free multilayer and analytic resonance for upward propagation becomes less valid. The
error also increases with defocus for large defects. As focus changes, the effect of the phase in
the predicted near field becomes more important. If small errors are made by ray tracing in the
phase prediction, these are amplified when the aerial image is calculated out of focus.
The effect of defect phase through focus is seen in the results for RADICAL with SSA.
For the case with no defocus, RADICAL SSA matches FDTD almost exactly for the largest
defect. But for the out of focus cases it over predicts the effect of the defect. This is because
SSA models the defect as purely a phase perturbation. Since RADICAL with SSA over predicts
the CD change, it suggests that in reality a buried defect affects the phase and magnitude of the
reflected field, not just the phase.
Since the error was expected to get worse for large defects and out of focus, as explained
above, the accuracy of RADICAL, while not perfect, is predictable. For simulations at best
focus and for defects smaller than 2.5nm surface height, RADICAL is accurate to about 1nm.
Outside of these cases, the accuracy is worse, but RADICAL will still provide useful insight into
the defect feature interaction.
Simulation results for the second scenario, a production EUV tool printing 22nm lines, are
shown in Figure 3-3. In this case, FDTD is not used because its accuracy is questionable for
22nm half-pitch patterns [26]. Instead, standard RADICAL is run on smoothed and un-
smoothed multilayers with the same surface defect geometry and the results are compared to
RADICAL with SSA. The method which predicted the highest CD change in all cases is a full
RADICAL with ray tracing simulation of substrate defects smoothed to form shorter and wider
surface defects. The other two methods use only the surface profile information. One is a full
RADICAL with ray tracing simulation of a defective multilayer with identical layers from the
surface to the substrate, the other is a RADICAL with SSA simulation. The uniform layer and
SSA simulations use the increased surface height approximation described in Section 4.2.4.
Figure 3-3. CD change as a function of surface defect height for three different geometry assumptions. Each defect
is located in the center of a 22nm dense line space pattern (wafer scale). NA=0.32, σ=0.75.
0 0.5 1 1.5 2 2.5-5
0
5
10
15
20
Surface Defect Height (nm)
CD
Change (nm
)
Focus = -75nm
0 0.5 1 1.5 2 2.5-5
0
5
10
15
20
Surface Defect Height (nm)
CD
Change (nm
)
Focus = 0nm
0 0.5 1 1.5 2 2.5-5
0
5
10
15
20
Surface Defect Height (nm)
CD
Change (nm
)
Focus = 75nm
Ray Tracing with Smoothing
Ray Tracing with Uniform Multilayer
Single Surface Approximation
21
The match between RADICAL with SSA and standard RADICAL with ray tracing is
clearly much better in Figure 3-2, for 32nm lines and NA=0.25 than in Figure 3-3, for 22nm
lines and NA=0.32. This difference is caused by the non-uniform layers below the surface,
which have a larger effect on the final image for tighter pitches and higher NA. The 22nm
pattern diffracts more light at higher angles than the 32nm pattern and more of the light
diffracted at these higher angles is collected by the higher NA. These higher angles are
important because for light incident above the nominal angle the multilayer was designed for, the
resonance is diminished and light propagates deeper into the multilayer. Therefore, the shape of
the lower layers contributes more to the final reflection off the multilayer and the reflections off
of each layer are not simply added in phase. Therefore, the assumptions behind the single
surface approximation are less accurate. But, as shown in Figure 3-3, if the multilayer is uniform
SSA matches the ray tracing method very well because each of the lower layers probed by the
off-axis incident light is identical to the top layer.
The dependence on angle is best demonstrated by comparing the phase of electric fields
reflected from an EUV mask at two angles of incidence, as shown in Figure 3-4. There is a 15%
change in the reflected phase for the off-axis light compared to the nearly on-axis light. For the
SSA approximation these two profiles would be identical, which is a source of error for SSA.
SSA is a fast and accurate method for some defects, but it may not be accurate enough for
absorber patterns with half-pitches below 32nm (wafer scale) on top of a multilayer with
significant non-uniformities below the surface.
Figure 3-4. Phase of reflected field cutline for two angles of incidence for a full ray tracing simulation of a 1.5nm
tall surface defect resulting from a 5nm tall substrate defect.
It is important to note that RADICAL assumes the layers in the mask continuous. The
accuracy of the approximations in SSA and the ray tracing method will be very bad if an abrupt
disruption in the multilayer occurs near the surface. This could be caused by a crack, or a large
defect particle added during the multilayer deposition in a layer near the top. For these cases a
rigorous simulation will be necessary.
3.2.3.1 Accuracy of Absorber Model for Corners RADICAL must be able to simulate more than just line space patterns, it must be able to
accurately simulate an arbitrary geometry. A simulation study was performed by Samsung to
investigate the printability of absorber defects [42]. These results are suitable for comparison
0 50 100 150 200 250 300 350 400 450 500-0.5
0
0.5
1
1.5
2
Distance (nm)
Phase (ra
dia
ns)
Phase comparison for two angles of incidence
1.5deg
10.3deg
22
with RADICAL simulations to verify the accuracy of RADICAL’s edge model. The Samsung
data used the waveguide simulation method, which is a rigorous method which works very well
for simulating EUV masks without buried defects.
Figure 3-5. Geometries simulated by Samsung and RADICAL.
The geometries simulated are summarized in Figure 3-5. The absorber defects are squares
with an increasing width, w. The maximum critical dimension change at the wafer due to the
defect was measured and the results are plotted in Figure 3-6. It is clear from Figure 3-6 that
RADICAL matched the rigorous data from Samsung very well, confirming that RADICAL can
accurately simulate absorber corners.
Figure 3-6. Comparison of CD change (1x wafer scale) predicted by rigorous simulation results from Samsung and
RADICAL as a function of absorber defect square width (4x mask scale). NA=0.25, σ=0.5.
Mask Distance (nm)
Mask D
ista
nce (nm
)
0 200 400 600
0
100
200
300
400
500
600
700
W
W
6°
0 10 20 30 40 50 60 70 80
0
5
10
15
20
25Comparison to Samsung Data
Square Width: w (nm)
Absolu
te V
alu
e o
f C
D C
hange (nm
)
RADICAL
Samsung Data
23
3.2.4 Computational Requirements of RADICAL
RADICAL is much less computationally intensive than FDTD. Table 3-1 shows the
runtime of FDTD, standard RADICAL, and RADICAL with SSA. These runtimes are for a
mask with a 128nm line space pattern and a buried defect. 128nm on the mask corresponds to
32nm on the wafer, assuming a 4x system. The total mask area simulated is 256nm x 256nm.
The buried defect is a circularly symmetric Gaussian 20nm tall and 50nm full width half max
(FWHM). The profiles of the 40 bilayers between the substrate and surface are predicted by the
model in [43]. The resulting 3.1nm tall and 61nm FWHM surface defect is centered on the edge
of the absorber. The FDTD simulator used is TEMPEST 6.0 with 30 nodes per wavelength.
FDTD is used as the rigorous method for comparison, instead of FEM or a modal method,
because it was available for free and its runtime is a predictable function of cell density and does
not depend and a program specific algorithm used to form rectangles or triangles to model the
defective multilayer geometry. To ease comparison with other methods, and between the
multiple computers used for this work, runtime values will be reported in an arbitrary unit. This
arbitrary unit is the time it takes to run the fft2 function in MATLAB on a 1000x1000 cell
matrix. On a 2.39GHz laptop, the unit is 0.109s.
Standard RADICAL is 980 times faster than FDTD for the line space pattern and
RADICAL with SSA is 25 times faster than standard RADICAL. This makes RADICAL with
SSA nearly 25,000 times faster than FDTD.
One interesting characteristic of RADICAL is the dependence of simulator runtime on
absorber pattern. This is a result of Fourier transform link between the absorber and multilayer
simulators within RADICAL. If the pattern is a simple line space pattern then only a few orders
of the Fourier transform along a single axis in k-space are needed to represent the pattern and
therefore only a few different incident plane waves need to be simulated by the ray tracing
simulator. But, if a more complicated pattern is simulated, like a contact or elbow, then many
more orders, and in turn multilayer simulations, are required. The runtime for FDTD simulations
depends only on the number of simulation nodes, not the pattern.
For standard RADICAL, an elbow pattern takes 2.4 times longer than a line space pattern.
However, in the SSA version the elbow pattern only takes 35% longer. This is explained by the
runtime distribution of the two versions. Figure 3-7 shows the runtime distributions for 128nm
line space patterns whose runtimes are shown in Table 3-1
2200
920
Standard
901,892
901,892
TEMPEST
RADICAL
502D Pattern
(Elbow)
371D Pattern
(Lines)
with SSA
2200
920
Standard
901,892
901,892
TEMPEST
RADICAL
502D Pattern
(Elbow)
371D Pattern
(Lines)
with SSA
Table 3-1. Runtimes of TEMPEST, RADICAL and RADICAL with SSA for a 256nm x 256nm mask
area. The unit is the time it takes to run the fft2 function in MATLAB on a 1000x1000 cell matrix
24
Figure 3-7 shows that for standard RADICAL the runtime is dominated by the multilayer
ray tracing simulations. Therefore, if the number of ray tracing simulations is scaled by some
factor, then the total runtime is scaled by approximately that same factor. For RADICAL with
SSA however, the SSA multilayer simulations are only a small fraction of the total runtime. So
if the number of SSA simulations required is doubled or tripled, it will only cause a few percent
increase in total runtime.
RADICAL also requires much less memory than FDTD. For an accurate simulation of
EUV masks with buried defects, FDTD needs a finer grid than would be necessary for a standard
DUV photomask simulation of the same area. This is due to numerical dispersion. It was shown
in [26] that for incident angles above 10°, the reflection coefficient begins to vary significantly as
a function of simulation grid size. The nominal mask incident angle in EUV lithography is 6°
and disruption of the multilayer by the defect will cause effectively higher angles of incidence
for some locations in the multilayer. This means that FDTD will need a high enough cells per
wavelength value to be accurate for angles above 10°. For this analysis, 40 nodes per
wavelength will be assumed as the cell density required for FDTD. Forty nodes per wavelength
corresponds to a dx value of 0.337nm. An EUV mask simulation domain is approximately
400nm tall. This is determined by the physical EUV multilayer and absorber height. The size of
the matrix needed to store the geometry for FDTD, assuming an Xnm x Xnm mask area, is given
by
Equation 3-1: ( )nodesX
nm
nmX
dx
heightlengthwidthN 42
3
2
310
337.
400⋅≈
⋅=
⋅⋅=
where X is in nm.
For standard RADICAL, the entire multilayer geometry must be stored. But, instead of a
dense grid through all of space, only the locations of material interfaces are stored. This means
that for a multilayer with 40 bilayers, 81 matrices would be needed; two for each bilayer and one
for the capping layer. Also, since the ray tracing algorithm doesn’t suffer from numerical
7%
Propagate
Up Through
Absorber
Perform
SSA
Simulations
Propagate down
through absorber
Standard RADICAL RADICAL with SSA
Perform
Ray Tracing
Simulations
7%
Propagate
Up Through
Absorber
Perform
SSA
Simulations
Propagate down
through absorber
Standard RADICAL RADICAL with SSA
Perform
Ray Tracing
Simulations
Figure 3-7. Runtime distribution for standard RADICAL and RADICAL with SSA
25
dispersion, it doesn’t require the dense grid that RADICAL does; a dx value of 1nm is sufficient.
The size of the matrix needed to store the geometry for standard RADICAL, assuming an Xnm x
Xnm mask area, is given by:
Equation 3-2: ( )nodesX
nm
X
dx
nlengthwidthN
layers100
1
81 2
2
2
2⋅≈
⋅=
⋅⋅=
where X is in nm.
For RADICAL with SSA only the geometry of top layer of the multilayer needs to be
stored. The size of the matrix needed to store the geometry for standard RADICAL with SSA,
assuming an Xnm x Xnm mask area, is given by:
Equation 3-3: ( )nodesX
nm
X
dx
lengthwidthN 2
2
2
21
≈=⋅
=
where X is in nm.
This analysis shows that for all methods the memory required scales as a function of mask
area simulated. Because RADICAL with SSA ignores the effects of the layers below the surface
it uses about 100 times less memory than standard RADICAL with the ray tracing simulator for
the multilayer. The ray tracing simulator must only store the boundaries between each layer.
Therefore is uses about 100 times less memory than FDTD, which must store a material value
for every simulation point. For example, 100MB of memory could simulate an area of 25µm2
using RADICAL with SSA, 0.25µm2 using standard RADICAL and only 0.0025 µm
2 with
FDTD.
3.3 Summary
The fast simulator RADICAL is able to accurately simulate the reflection from an EUV
mask with a buried defects much more efficiently than FDTD. When the ray tracing method is
used, it’s nearly 1000 times faster and uses 100 times less memory. When the advanced SSA
method is used it is nearly 25,000 times faster than FDTD and uses 10,000 times less memory.
The rest of this dissertation will cover applications of RADICAL to EUV lithography research
and development.
26
4 Printability of Isolated
Defects This chapter will focus solely on the electromagnetic properties and effects on lithographic
imaging of defects that are not near features, referred to as isolated defects. The results of the
work in this chapter, specifically the through focus behavior of an isolated defect image, will be
important in every chapter of this dissertation.
First the phase nature of defects will be demonstrated through simulation by comparisons
to other types of defects. Then, the role of multilayer smoothing will be investigated and a
simple model for the defect’s aerial image will be introduced along with improvements to the
single surface approximation. Next, the effects of illumination and vertical defect position
within the multilayer stack position will be investigated. Finally, RADIAL will be compared to
actinic inspection images to confirm the conclusions from simulation and demonstrate the
accuracy of RADICAL.
4.1 Phase Nature of Defects
It is common for people in the EUV community to refer to buried defects as phase defects.
This can be a misleading name because buried defects affect both the phase and magnitude of the
reflected field. But, it is true that the dominant effect of buried defects is a phase effect. This is
shown in Figure 4-1, which is a plot of the center intensity of the aerial image of four different
defects through focus. These defects were carefully chosen to have the same full width half max
(FWHM) and center intensity at best focus. Each defect has a FWHM of about 60nm and is
normalized to a defect free background level of 1.0. A FWHM of 60nm is chosen because it is
close to the sizes of the defects created on the Intel programmed defect mask described in Table
2-1. The thin mask amplitude defect, which assumes the defect is square and has a transmission
of zero, has an aerial image center intensity of about 0.55. The required heights of the actual
buried defects and phase of the thin mask phase defect were determined by iteration. Figure 4-1
is summary of the parameters for all four defects.
Table 4-1. Summary of defects simulated in Figure 4-1. The height and width are defined on the surface of the
multilayer.
Defect Type FWHM (nm) Height (nm) Phase (degrees) Transmission
(Fraction of Incident)
Real Bump 60 4.0 not applicable not applicable
Real Pit 60 -5.5 not applicable not applicable
Thin Mask
Amplitude 60 not applicable not applicable 0
Thin Mask Phase 60 not applicable 100 1.0
Figure 4-1 shows that the ideal amplitude defect has its lowest intensity when it is in focus.
The ideal phase defect, however, has the lowest intensity out of focus and the intensity inverts
through focus. The buried defects, like ideal phase defect, have their worst case intensity out of
focus. But, they cannot simply be modeled as ideal phase defects because even though the
27
intensity value is the same in focus, the swing through focus is much larger for the buried
defects. It is also interesting to note that although the pit defect has the same FWHM on the
mask as the bump defect, it requires a larger surface height to produce the same center intensity
at best focus. But, through focus its worst case intensity is lower than the worst case for the
bump defect. Also, worst case for a pit occurs at positive focus and the worst case for a bump is
at a negative focus value.
Figure 4-1. Center intensity of the aerial image of four types of defects. NA=0.32 σ=0.75.
4.2 Simple Model for Isolated Defects
Although a simple thin mask constant phase defect model does not correctly predict the
image intensity behavior of a buried defect through focus, if a constant smoothing model is
assumed there is a simple relationship between the surface height and resulting image of the
isolated defect that holds for a wide range of defect shapes and sizes. The development of this
model will be presented in this section. First, the effects of multilayer smoothing will be studied
by using simulation to artificially control the smoothing process to hold certain geometric
properties of the multilayer constant while varying others. Then, a realistic smoothing model
will be used to develop an algebraic defect image intensity model.
4.2.1 Introduction to Smoothing
Multilayer smoothing during deposition has been shown to effectively reduce printability
of substrate defects. When multilayer smoothing is applied during deposition, the height of the
defect is reduced for each layer deposited. So, after 40 bilayers have been deposited the resulting
surface height of the defect is much less than the substrate height. Smoothing has a more
complicated effect on surface defect width. For the model discussed in this section, the
smoothing process causes the surface width to be relatively independent of substrate defect
width. A mathematical model has been developed to describe this growth process [43]. This
model will be used, with the specific input parameters varied, for all the simulations in this
section.
-250 -200 -150 -100 -50 0 50 100 150 200 2500
0.2
0.4
0.6
0.8
1
Focus (nm)
Cente
r Im
age Inte
nsity
Center Intensity of Aerial Image Through Focus
Real Buried Bump:: 4nm Surface Height
Real Buried Pit: -5.5nm Surface Height
Thin Mask Amplitude Defect
Thin Mask Phase Defect
28
Figure 4-2 shows a set of growth results, surface defect height and width as a function of
buried defect size, from a programmed defect mask fabrication experiment and simulation. The
smoothing parameters used for these simulations were tuned to match the experimental growth
results. The experimental mask is the Intel programmed defect mask described in Section 2.4.4.
The programmed substrate defects each have a constant height of 48nm, and square bases with
widths that varied from 60nm to 90nm. The results in Figure 4-2 show that the final surface
defects all have relatively constant widths of about 50-60nm but heights that vary from 1.5nm to
5.5nm. This is surprising because the heights of the buried defects were constant and the widths
were varied. These results demonstrate an important feature of this smoothing model: surface
defects from arbitrary buried defect sizes have fairly constant widths. The surface height is what
is determined by the buried defect size and shape. The values of the smoothing parameters
defined in [43] which were used to match the programmed defect mask results are summarized
in Table 4-2. The two parameters adjusted to control the surface defect size for a given substrate
defect were “Si etched per bilayer”, which primarily effected the surface height and $%dep
which
primarily affected the surface width.
Figure 4-2. Comparison of experimental and simulated growth results for a defect with buried a height of 48nm and
a square base with varying width.
Table 4-2. Multilayer growth parameters used to match Intel programmed defect mask
Growth Parameters Value
Number of bilayers (N) 40
Bilayer Thickness (nm) 7.0
Contraction per bilayer (nm) 0.8
Si etched per bilayer (nm) .91
$)
dep (unitless) 0
$%
dep (nm) 0.47
$*
dep (nm2) 0
$+
dep (nm3) 6.0
$)etch (unitless) 0.3
$%etch (nm) 0
$*etch (nm2) 0
$+etch (nm3) 0
60 65 70 75 80 85 900
1
2
3
4
5
6
Buried Width (nm)
Surface H
eig
ht (n
m)
Surface Height as a Function of Buriend Width for Buried Height=48nm
Berkeley Sim
Intel Experiment
60 65 70 75 80 85 900
1
2
3
4
5
6
Buried Width (nm)
Surface H
eig
ht (n
m)
Surface Height as a Function of Buriend Width for Buried Height=48nm
Berkeley Sim
Intel Experiment
60 65 70 75 80 85 900
10
20
30
40
50
60
70Surface Width as a Function of Buriend Width for Buried Height=48nm
Surface W
idth
(nm
)
Buried Width (nm)
60 65 70 75 80 85 900
10
20
30
40
50
60
70Surface Width as a Function of Buriend Width for Buried Height=48nm
Surface W
idth
(nm
)
Buried Width (nm)
29
4.2.2 Analysis of Smoothing
Figure 4-3. Definition of dip strength
The flexible model defined in [43] allows interesting and physically impossible
simulations to be performed. These simulations allow the effects of certain parameters of the
multilayer to be isolated to determine what is important for a fast model. For example, the
smoothing model parameters can be varied to force a constant surface defect size for varying
buried defect sizes. Also, for constant buried defects the model parameters can be adjusted to
produce different surface defects. The reflected fields and aerial images produced by these two
sets of geometries give important insight into what geometrical aspects are important. All
simulations were done in three-dimensions. The geometries shown in the figures are cutlines.
The Gaussian defects are rotationally symmetric. The box defects, referred to as “tetra,” have
square bases.
The next few sections of this chapter will artificially create defect geometries and compare
their printability. This will help to isolate which geometric properties of the defective multilayer
have the largest effect on the final aerial image. All aerial image simulations in Section 4.2 are
for 0.25NA, partial coherence (σ) of 0.75 and are at best focus.
4.2.2.1 Constant Surface Geometry Defect Simulations Figure 4-4 shows two multilayer geometries with the same size surface defect FWHM and
height, but different buried defects. Both geometries have a surface defect with a height of
3.86nm and a full width half max (FWHM) of 67.5nm. The smoothing process parameters were
adjusted by an iterative method to produce these geometries. The left geometry in Figure 4-4 has
a Gaussian buried defect with a height of 10nm and a FWHM of 50nm, the right geometry has a
height and FWHM of 50nm. The resulting aerial image dip strengths are plotted in Figure 4-5.
The dip strength is defined in Figure 4-3.
0 50 100 150 200 250 300 350 400 450 5000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Icenter
Ibackground
background
centerbackground
I
IIDip
−=
30
Figure 4-4 Multilayer geometries with smoothing model parameters adjusted to force a constant surface defect.
Both the results from the ray tracing simulator [26] and the single surface approximation
[36] are shown, the differences between the two results are discussed below. The dip strength of
the aerial image is nearly constant for the constant surface defects. This is a surprising result
considering the apparent differences in the lower layers of the multilayer geometries shown in
Figure 4-4. It suggests that only the top surface of the multilayer determines that reflection.
This is an encouraging result for modeling and inspection because if only information from the
top layer is needed then a reduced model that doesn’t require the full defective mask blank
geometry is possible. Upon closer inspection of the multilayer geometries in Figure 4-4, the
results in Figure 4-5 are not that surprising. Most of the light reflected from an EUV mask does
not penetrate very deeply into the multilayer stack so only the top few layers are important for
predicting the reflected field. In these two geometries, the layers near the top are nearly
identical.
50 100 150 200 250 300 350 400 450 5000
50
100
150
200
250
300
Heig
ht (n
m)
50 100 150 200 250 300 350 400 450 5000
50
100
150
200
250
300
Heig
ht (n
m)
50 100 150 200 250 300 350 400 450 5000
50
100
150
200
250
300
X Distance (nm)
Heig
ht (n
m)
50 100 150 200 250 300 350 400 450 5000
50
100
150
200
250
300
X Distance (nm)
Heig
ht (n
m)
X Distance (nm)X Distance (nm)
5 10 15 20 25 30 35 40 450
0.2
0.4
0.6
0.8
1
Buried Defect Height (nm)
Dip
Strength
(I b
ackg
round-Icente
r)/Ibackg
round
Aerial Image Dip Strength for Constant Surface Defect Height 3.86nm and (FWHM): 68nm
Ray Tracing
SSA
Figure 4-5. Aerial image dip strength as a function of buried
defect height for a constant surface defect size. NA=0.25 σ=0.75
31
Figure 4-6. Multilayer geometries with smoothing model parameters adjusted to force produce different surface
defects for the same buried defect
3.4 3.6 3.8 4 4.2 4.4 4.6 4.8 5 5.2 5.40
0.2
0.4
0.6
0.8
1
Surface Defect Height (nm)
Dip
Strength
(I b
ackg
round-Icente
r)/Ibackg
round
Aerial Image Dip Strength for Constant Buried Gaussian Defect Height 30nm and (FWHM): 50nm
Ray Tracing
SSA
Figure 4-7. Aerial image dip strength as a function of surface defect height for a constant buried defect size.
NA=0.25 σ=0.75
4.2.2.2 Constant Buried Defect Geometry Simulations Figure 4-6 shows two multilayer blank geometries with identical buried Gaussian defects
30nm tall and 50nm wide, but different surface defects. The smoothing model parameters used
are two cases from the constant surface defect simulations above. The top geometry was
produced by etching 0.24nm of Si for each bilayer deposition and the bottom geometry was
created by etching 1.18nm. The geometries look very similar. The upper geometry has a surface
defect height of 3.28nm and the lower geometry has a surface defect height of 5.47nm. Figure
4-7 shows the resulting dip strength of the aerial image from these geometries and others with a
constant buried defect. The minimum dip strength is 0.5 and the maximum is 0.8. This is a
much more significant increase than that seen in Figure 4-5 for the constant surface defect case.
Although the geometries in Figure 4-6 look very similar, discovering they produce very different
images is not very surprising. The minimum defect height is 24% of the 13.5nm wavelength and
the maximum defect height is 41% of the wavelength. The interference caused by phase defects
of this size relative to the wavelength should be different due to the different phases of the
reflected orders created by the defect. These constant buried defect geometry results support the
evidence from the constant surface geometry defect simulations that suggest that the surface
defect profile is what affects printing. Physically small defects on the multilayer surface caused
non-trivial variation in the resulting aerial image.
50 100 150 200 250 300 350 400 450 5000
50
100
150
200
250
300
Heig
ht (n
m)
X Distance (nm)
50 100 150 200 250 300 350 400 450 5000
50
100
150
200
250
300
X Distance (nm)
Heig
ht (n
m)
X Distance (nm)X Distance (nm)
32
Figure 4-8 gives even more insight into what parameters are important for a fast model. It
shows the reflected phases for the two groups of simulations shown above. For the constant
defect case, the reflected phase is virtually unchanged between the simulations. For the constant
buried defect simulations, there is significant variation between the simulations. This variation
in phase corresponds to the variation in dip strength shown above. This suggests that predicting
the reflected phase is critical to predicting the aerial image dip. If this is true, than the single
surface approximation (SSA) model proposed by Gullikson in [36] should be a good model for
masks with buried defects. This will be addressed later in this chapter.
4.2.2.3 Constant Smoothing Process Simulation
The two sets of simulations above used artificial smoothing model parameters to isolate the
effects of the surface and lower layers of the mask. To develop a useable model, more realistic
simulations are needed. The goal of the following simulations is to develop a very fast model for
defect printability. The previous simulations have suggested that by using only the surface
information, the dip strength can be predicted. This set of simulations will be used to quantify
that into a new fast model. Figure 4-2 shows actual experimental data along with simulation
results calibrated to match the data. The following simulations were performed using these
X Distance (nm)
0 100 200 300 400 500-2
0
2
4
6
X Distance (nm)P
hase o
f N
ear Fie
ld (ra
d)
5.47nm
3.28nm
Constant Surface Defect Constant Buried Defect
0 100 200 300 400 500-2
0
2
4
6
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
X Distance (nm)
0 100 200 300 400 500-2
0
2
4
6
X Distance (nm)P
hase o
f N
ear Fie
ld (ra
d)
5.47nm
3.28nm
Constant Surface Defect Constant Buried Defect
0 100 200 300 400 500-2
0
2
4
6
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
Figure 4-8. Phases of reflected nearfields for the two types of defective blank geometries simulated. Labels
correspond to surface defect height
0 0.5 1 1.5 2 2.5 3 3.5
x 105
0
2
4
6
8
10
Buried Defect Volume (nm3)
Surface H
eig
ht (n
m)
Surface Height
0 0.5 1 1.5 2 2.5 3 3.5
x 105
0
20
40
60
80
100
Buried Defect Volume (nm3)
Surface W
idth
(nm
)
Surface Width
Gauss_w75
Gauss_w50
Tetra_w50Tetra_w75
Gauss_w75
Gauss_w50
Tetra_w50
Tetra_w75
Figure 4-9. Surface defect height and width as a function of buried defect volume. The labels correspond
to the shape and width of the buried defect. The resulting height and width appears to be a predictable
function of height for a given buried shape and width, but this function is different for different buried
shapes and widths.
33
smoothing model parameters for many different sizes and shapes of substrate defects. The
results of the smoothing simulations are shown in Figure 4-9.
Figure 4-9 shows the complexity of the smoothing process. The surface defect size
depends on the buried defect’s shape and size and there does not appear to be any simple model
to predict the surface defect size. This strong dependence on shape means that quickly predicting
the growth, and resulting image, based on knowledge of the substrate defect only would be very
difficult.
The resulting simulated reflected electric fields for the all of the geometries shown in
Figure 4-9 are shown in Figure 4-10. These near field dip strength results are not a simple
function of surface defect height. The four branches in the dip strength plot, corresponding to
the four branches in the plots in Figure 4-9, suggest that the more complicated lower layer
geometries, not just the surface profile, affect the reflected electric field magnitude. This makes
a simple model for the near field dip strength impossible.
4.2.3 Algebraic Isolated Defect Model
Figure 4-11 shows the resulting aerial images computed from the electric fields in Figure
4-10. These results show the same single parameter dependence as the constant buried and
constant surface defect cases. The major reason for the single parameter dependence is the
smoothing process. Due to the smoothing process, the final height and width are correlated and
the shape of the top surface of the multilayer is independent of buried defect shapes. Also, the
surface profile is always very close to Gaussian for any buried defect geometry. This single
parameter dependence means a simple and fast model for the aerial image is possible.
It is important to note that the aerial image drip strength could be predicted by a single
parameter, even though the nearfield could not be. The aerial image can be roughly thought of
as a low-pass filtered version of the near field intensity. If low-pass filtering reduces the effects
of the lower layers, it suggests that these lower layers produce reflections at higher angles which
do not affect the image at this numerical aperture (NA), but could at higher NA.
0 1 2 3 4 5 6 7 8 9 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Surface Defect Height (nm)
Dip
Strength
: (b
ackgro
und-c
ente
r)/b
ackgro
und)
Near Field Dip Strength
0 50 100 150 200 250 300 350 400 450 5000.2
0.4
0.6
0.8
1
Reflected Near Field
X Distance (nm)
Magnitude N
ear Fie
ld(E
)
0 50 100 150 200 250 300 350 400 450 500-5
0
5
10
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
0 1 2 3 4 5 6 7 8 9 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Surface Defect Height (nm)
Dip
Strength
: (b
ackgro
und-c
ente
r)/b
ackgro
und)
Near Field Dip Strength
0 50 100 150 200 250 300 350 400 450 5000.2
0.4
0.6
0.8
1
Reflected Near Field
X Distance (nm)
Magnitude N
ear Fie
ld(E
)
0 50 100 150 200 250 300 350 400 450 500-5
0
5
10
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
Figure 4-10. Reflected magnitude, phase, and dip strength of the magnitude for the electric fields reflected from
EUV mask blanks created with experimentally based smoothing parameters
34
Figure 4-12 Linear portion of aerial image dip strength from Figure 4-11 with linear fit.
While the surface height itself depends on the buried defect size and shape as well as the
smoothing process used during multilayer deposition, Figure 4-12 shows that the dip strength of
the aerial image at best focus can be approximated by a linear model for small defects that
depends only on the surface defect height. This means that the results of a simulation that would
take days with TEMPEST and minutes with RADICAL can now be predicted instantly. The
following formula gives the dip strength as a function of surface defect height.
Equation 4-1:
4.2.4 Tuned Single Surface Approximation
The algebraic model above is more accurate than the single surface approximation, though
it actually requires less information than SSA. Unlike the single surface approximation, this
model has no physical basis; it is just a curve-fit of simulation results. The single surface
approximation on the other hand has real physical meaning. It turns out that SSA can be
modified slightly to produce better results while still being physically based.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8Linear Approximation of Aerial Image Dip Strength
Surface Defect Height (nm)
Dip
Strength
(I b
ackg
round-Icente
r)/Ibackg
round
094.0191.0)( 1 −⋅= − hnmhDip
0 50 100 150 200 250 300 350 400 450 5000.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Aerial Image Cutlines
Aerial Im
age Inte
nsity (E2)
X Distance (nm)
1 2 3 4 5 6 7 8 90
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Surface Defect Height (nm)
Dip
Strength
(I b
ackg
round-Icente
r)/Ibackg
round
Aerial Image Dip Strength for Constant Smoothing Model
Figure 4-11. Reflected aerial image and aerial image dip strength from EUV mask blanks created
with experimentally based smoothing parameters. NA=0.25 σ=0.75
35
In Figure 4-5, Figure 4-7 and the left plot of Figure 4-14 below, it is clear that SSA
consistently under predicts the printability of buried defects. The reason for the error is shown
clearly in Figure 4-13. On the left, the reflected phases from the standard SSA model and ray
tracing simulator are shown. The SSA phase change due to the defect is less than the actual
phase change calculated by ray tracing. This is an important result because it shows that even
though the reflected phase is determined mainly the surface defect profile, it is not just a phase
correction based on path difference as SSA assumes. There is actually a more pronounced effect
due to the lower layers in the multilayer. Since, due to smoothing, each layer has a slightly
lower defect height than the defect below it, using a lower layer as the “single surface” in SSA
produces more accurate results. For this smoothing model it turned out that using the third layer
from the top makes the results of SSA most accurate. Because the light reflected does propagate
down to layers below the top surface, using a lower layer as an effective or average single
surface is a physically reasonable model. The right plot in Figure 4-13 shows that using this
lower layer produces a more accurate reflected phase. The right plot in Figure 4-14 shows that
for small defects this modified SSA becomes more accurate than the standard SSA method.
Small defects, less than 4nm on the multilayer surface, are assumed to be the most important
because as substrate quality and smoothing methods improve large defects will be rare, but small
defects that still affect printing will remain.
It’s important to note that in [36] SSA was introduced and used with a separate
approximate method to predict the aerial image. Previous publications have used both SSA and
the approximate aerial image method and concluded that SSA was very inaccurate [37]. Figure
4-14 shows that when a more rigorous and accurate aerial image simulator is used, the near fields
generated by SSA are fairly accurate, even without the lower surface modifications.
Also, the accuracy variation of SSA with varied smoothing is shown in Figure 4-7. As the
surface defect size increases, the effects of smoothing decrease. Therefore, the top surface layer
profile is a better approximation for the lower layer profiles, making SSA more accurate
0 50 100 150 200 250 300 350 400 450 500
0
1
2
3
4
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
SSA Phase
Ray Tracing Phase
0 50 100 150 200 250 300 350 400 450 500
0
1
2
3
4
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
SSA Phase
Ray Tracing Phase
0 50 100 150 200 250 300 350 400 450 500
0
1
2
3
4
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
SSA Phase
Ray Tracing Phase
0 50 100 150 200 250 300 350 400 450 500
0
1
2
3
4
X Distance (nm)
Phase o
f N
ear Fie
ld (ra
d)
SSA Phase
Ray Tracing Phase
Figure 4-13. Comparison of phase of reflected field between SSA and ray tracing from a 3.2nm tall surface
defect. Left: Standard SSA Model. Right: Modified SSA model using lower layer
36
1 2 3 4 5 6 7 8 90
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Surface Defec t Height (nm)
Dip
Strength
(I b
ackgro
und-Icente
r)/I b
ackg
round
Aerial Image Dip S trength for Cons tant Sm oothing M odel
Ray Trac ing
M odified SSA
1 2 3 4 5 6 7 8 90
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Surface Defec t Height (nm )
Dip
Strength
(I b
ackg
round-Icente
r)/I b
ackg
round
Aerial Im age Dip S trength for Cons tant Smoothing Model
Ray Trac ing
SSA
1 2 3 4 5 6 7 8 90
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Surface Defec t Height (nm)
Dip
Strength
(I b
ackgro
und-Icente
r)/I b
ackg
round
Aerial Image Dip S trength for Cons tant Sm oothing M odel
Ray Trac ing
M odified SSA
1 2 3 4 5 6 7 8 90
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Surface Defec t Height (nm )
Dip
Strength
(I b
ackg
round-Icente
r)/I b
ackg
round
Aerial Im age Dip S trength for Cons tant Smoothing Model
Ray Trac ing
SSA
4.3 Effects of Illumination
The algebraic isolated defect model is accurate only for predicting the dip strength of the
image of an isolated defect for a certain optical system. This is because the image of an isolated
buried defect is strongly dependent on the illumination. Figure 4-15 shows the center intensity
of a buried defect for the three types of illumination, with the dipole and annular illuminations
optimized for printing 16nm line space patterns. The inversion of the center intensity does not
occur with annular and dipole illumination. Also, the minimum intensity, around -75nm
defocus, does not appear to be as bad for these illuminations as it is for tophat. Although the
center intensity values are nearly identical for annular and dipole illuminations, it is important to
note that the images are quite different, as shown in Figure 4-16. This suggests that for
illuminations other than tophat, the center intensity of an isolated defect image is not a reliable
metric to judge the impact of a defect.
Figure 4-15.Center intensity of the aerial image a buried defect as a function of wafer focus level for three
illuminations
-200 -150 -100 -50 0 50 100 150 2000.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Focus (nm)
Cente
r Im
age Inte
nsity
Center Intensity of Aerial Image Through Focus
Tophat
Annular 16nm
Dipole 16nm
Background Intensity
Figure 4-14. Plots of aerial image dip strength for ray tracing and SSA as a function of surface
defect height. Left: Standard SSA method on top surface. Right: Modified SSA method using
lower surface
37
Figure 4-16. Image of isolated buried defect at +125nm defocus for three illuminations
4.4 Mask blank inspection
There are currently no commercially viable actinic, meaning at EUV wavelength,
inspection tools for EUV masks or mask blanks. There is some debate in the industry whether or
not actinic mask inspection tools are necessary, or if optical tools are good enough. If an actinic
blank inspection tool is not used, an existing tool with a longer wavelength would be used. Two
possible tools are the Lasertec M7360, which operates at a wavelength of 266nm [44], and the
KLA-Tencor Teron which operates at a wavelength of 193nm [45]. The major difference
between these tools and an actinic tool, is the penetration of the inspection light into the EUV
multilayer. The actinic light will penetrate many layers into the multilayer and the resulting
inspection image will be affected by many layers below the surface. The longer wavelengths,
however, will not penetrate into the multilayer and therefore the inspection image will be a result
of only the surface geometry.
The results in this chapter suggest that non-actinic light will be adequate for EUV mask
blank inspection. It was shown that the surface height is what determines the image of an
isolated defect. Therefore, all an inspection system needs to do is look at the surface profile, so
using a wavelength of light able to penetrate the multilayer is not necessary.
There two important caveats to this conclusion. The first is that all of this analysis
assumed a smoothing model that made each layer very similar to the layer below it. This meant
that the surface layers of all the simulated masks were very similar to the layers near the surface.
If a multilayer deposition scheme is used for which this is not true, the conclusion that surface
inspection is adequate may not be true.
The second caveat is that surface inspection will likely result in many false defects being
detected. Because EUV light penetrates into the multilayer, a defect that is only on the surface,
and not on each layer before the surface as well, may not be printable in an EUV scanner, but it
would be detected by a non-actinic tool.
Examples of defects that would appear identical in a surface inspection, but different in an
actinic inspection of wafer print are shown in Figure 4-17 and Figure 4-18. Figure 4-17 shows
three defects on the surface, in the middle, and on the substrate of a blank. The surface profile of
each is identical. Figure 4-18 shows the resulting images of isolated defects deposited on
different layers. The defect on bilayer one, which is basically a surface defect, has the smallest
dip while the substrate defect, which affects every layer has the largest dip. The defect on layer
one would be a false defect in most situations.
Tophat Annular Dipole
38
Figure 4-17. Example refractive index map of three different of 10nm tall defects. Each defect was deposited on a
different layer of the multilayer. The deposition model assumes the layers below the defect are perfect and above
the defect are uniform.
Figure 4-18. Aerial image cutlines for Si and Mo defects on various layers.
These sorts of false defects have been seen in experiments. For example, research by
LaFontaine [7] has shown defects that are clearly visible in scanning electron microscope (SEM)
images of the mask. But, no effect of the defect is visible in the resulting wafer image. This
observation is likely due to a defect on the surface of the multilayer with a defect free multilayer
below. Because all materials have a refractive index near unity at 13.5nm, a surface defect that
is visible in an SEM may not cause any visible change in the wafer image. This work also shows
the opposite effect: a defect that is invisible in an SEM image but causes an unacceptable change
in the resulting wafer printing. This is most likely due to a substrate defect causing a surface
defect that is too short to be visible in an SEM, but large enough to have a serious effect on the
wafer image.
4.5 Isolated Defect Experiments
The dependence of isolated defect printability on size and focus shown in the previous
sections through simulation was confirmed by experiments on the Actinic Inspection Tool (AIT).
Two buried defects were chosen from the Intel programmed defect mask (PDM) described
earlier, one with a surface height of 2nm and the other with a surface height of 5.3nm. The
results of the experiments and simulations are shown in Figure 4-19 and Figure 4-20.
The AIT is not aberration free [46]. A method to use isolated defect images through focus
to determine the aberrations present in the AIT is described in the next chapter. That method
was used in the RADICAL simulations here to determine the aberrations present in the system.
An astigmatism value of 0.2 waves RMS (root mean squared) and a spherical aberration value of
0.025 waves RMS was used in the simulations in this section. These are much higher aberration
Si Defect on bilayer 20
Mo Defect on bilayer 40
(substrate)
Mo Defect on bilayer 1
0 50 100 1500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Distance (nm)
Aerial Im
age Inte
nsity
Bilayer: 40
Bilayer: 35
Bilayer: 30
Bilayer: 25
Bilayer: 20
Bilayer: 15
Bilayer: 10
Bilayer: 5
Bilayer: 1
39
values than would be expected in a production quality lithography tool, but are not as high as the
aberrations found in the next chapter. The images in this section were taken in December 2008
whereas the images in the next chapter were taken in August 2007. In that 8-month period
significant improvements were made by the team at Lawrence Berkeley National Lab to reduce
the aberrations of the AIT.
Figure 4-19. Through focus aerial images of isolated buried defect resulting in a 2nm tall surface defect. The label
on each column is the mask focal position. The top row is of oversampled images from the AIT. The bottom row is
images calculated by RADICAL.
Figure 4-20. Center intensity of aerial images for defects measured on AIT and calculated by RADICAL.
Figure 4-19 shows good quantitative agreement between RADICAL and the experimental
data for images of the 2nm tall defect. The inversion of the center intensity is obvious. The
effects of astigmatism are present as well. One effect of astigmatism present is the rotation of
the defect image counter-clockwise through focus. The second effect is the saddle shape of the
image near best focus. This is obscured by the noise in the AIT image but is obvious in the
RADICAL simulation. The fact that all of these effects are present in the AIT images and
RADICAL simulations suggests that the near fields calculated in RADICAL and the aerial image
model parameters used in Panoramic EMSuite are accurate.
Figure 4-20 shows a quantitative comparison of the center intensity values calculated by
RADICAL and measured on the images from the AIT. The match is very good for the 2nm
defect through focus. The match is not quite as good for the 5.3nm defect for positive focus.
Orders reflected at higher angles contribute most to the changes in aerial image compared to the
-3.6µm -2.8µm -2.1µm -1.3µm -0.6µm 0.2µm 1.0µm 1.7µm 2.5µm 3.2µm
-3 -2 -1 0 1 2 30
0.5
1
1.5
2
Focus (um)
Cente
r Im
age Inte
nsity
Center Intensity of Aerial Image Through Focus
Simulated 2nm Bump
Simulated 5.3nm Bump
Experiment AIT 2nm Bump
Experiment AIT 5.3nm Bump
Background Intensity
40
in focus case. This is due to the orders passing near the edge of the pupil undergoing a larger
phase change than the background. That means that the differences between the AIT and
RADICAL results are due to higher angles. One possible explanation for the difference in light
reflected at higher angles is that the lower layers of the multilayer are not modeled correctly in
RADICAL. The ray tracing simulator in RADICAL assumes that each layer, while distorted, is
composed of one material and totally separate from the other layers. These distinct layers will
cause reflections at high angles, because the lower layers are affected more by the defect than
upper layers due to smoothing. In reality, the layers an EUV mask near the defect will not be
perfectly distinct. If RADICAL predicts more light scattered at high angles than a real EUV
mask actually produces it could explain the differences seen for positive focus in Figure 4-20.
Since the defect is smaller for the 2nm case, this effect would not be as bad, which explains why
the 2nm defect matches the actual AIT images better than the 5.3nm defect.
4.6 Summary
Studying isolated defects reveals a lot about their electromagnetic properties and effects on
lithographic imaging. The most important result in this chapter, which was demonstrated
repeatedly, is that buried defects primarily effect the phase of the reflected field. This produces
an inversion in the defect aerial image intensity through focus. The magnitude of the phase
change is determined primarily by the surface geometry of the defect. In fact, for defects with a
surface height less than 4.5nm, the center intensity of the aerial image is a linear function of the
surface height. The SSA model was also improved upon by accounting for the average
propagation into the multilayer of the incident light. The final section of this chapter compared
simulated RADICAL results to experimental AIT images to confirm the effects observed in the
simulations as well as the accuracy of RADICAL. The complexities of these comparisons,
specifically determining the aberrations in the AIT, are covered in detail in the next chapter.
41
5 Using Isolated Buried
Defects to Extract
Aberrations The Actinic Inspection Tool (AIT) at Lawrence Berkeley National Laboratory is the most
advanced actinic, meaning it uses EUV light, tool in the world for capturing the aerial image of
an EUV mask. However, because it employs a zone plate for imaging and does not have state of
the art alignment controls it is susceptible to significant aberrations. A novel method was
developed to use RADICAL to extract the aberrations present in the AIT from through focus
images of a buried EUV mask defect. The data in this chapter was collected in August 2007.
The numerical aperture (NA) was 0.0625 at the mask. This is the same mask NA as a 4x
reduction scanner with a wafer NA of 0.25. The wavelength employed by the AIT is 13.4nm.
The focus values refer to the mask focus position and the focus is adjusted by physically moving
the mask up and down. One Rayleigh Unit of defocus is 1.7µm.
The presence of the aberrations mentioned in the previous chapter is obvious in the images
of isolated buried defects through focus shown in Figure 5-1. The physical defects in the mask
are symmetric, but the AIT images rotate from an ellipse with its major axis pointing northeast at
-7.6µm defocus to an ellipse with its major axis pointing northwest for +7.7µm defocus. This is
caused by astigmatism, as will be shown below. Figure 5-2 shows the initial comparisons
between the center intensity of the buried defect images from RADICAL and the AIT. It is
obvious that both the absolute intensities and trends in intensity through focus do not match.
It will be demonstrated that the magnitude of the aberrations present in the AIT can be
extracted by matching the center intensity of through focus AIT images with simulations
modeling the aberrations. It turns out that each aberration affects a different feature of the center
intensity versus focus curve, so the extraction does not require the simulation of every possible
aberration combination. In fact, each aberration focus combination must only be simulated once,
with the other aberrations fixed.
Focus: -7.6nm
X Distance (um)
Y D
ista
nce (
um
)
0 0.5 1
0
0.5
1
Focus: -3.6nm
X Distance (um)
Y D
ista
nce (
um
)
0 0.5 1
0
0.5
1
Focus: 0.3nm
X Distance (um)
Y D
ista
nce (
um
)
0 0.5 1
0
0.5
1
Focus: 4nm
X Distance (um)
Y D
ista
nce (
um
)
0 0.5 1
0
0.5
1
Focus: 7.7nm
X Distance (um)
Y D
ista
nce (
um
)
0 0.5 1
0
0.5
1
Focus -7.6µm Focus -3.6µm Focus 0.3µm Focus 4µm Focus 7.7µm
Figure 5-1. AIT aerial image of an isolated buried defect with a surface height of 6.2nm and a FWHM of 58nm
42
5.1 Astigmatism
Figure 5-3 shows the simulated aerial image out of focus with increasing levels of
astigmatism. The higher the astigmatism the more elliptical and rotated the defect image
becomes. This rotation is similar to the rotation in the images in Figure 5-1. The addition of
astigmatism into the simulation also affects the center defect intensity through focus. Figure 5-4
shows this effect. As astigmatism is increased, the slope of the curve decreases and the
maximum and minimum intensity points become less extreme. An astigmatism value of 0.55
waves matches the slope of the experimental values best around zero defocus.
Astig =0 Astig =0.2 Astig =0.4 Astig =0.6 Astig =0.8
Figure 5-3. RADICAL simulation of a 6.2nm tall (surface height) buried defect -5µm out of focus with
varying astigmatism values. The units of astigmatism are waves.
-10 -8 -6 -4 -2 0 2 4 6 8 100
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Focus
Inte
nsity
exp: defect 1
exp: defect 2
exp: defect 3
sim: NA=0.0625
Focus (µm)
Center Intensity of Aerial Im
age
-10 -8 -6 -4 -2 0 2 4 6 8 100
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Focus
Inte
nsity
exp: defect 1
exp: defect 2
exp: defect 3
sim: NA=0.0625
Focus (µm)
Center Intensity of Aerial Im
age
Figure 5-2. Comparison of center image intensity between RADICAL simulations and experiments. The
solid line is the RADICAL simulation with no aberrations included and the dotted lines are the experiment.
43
5.2 Spherical
Astigmatism is not the only aberration present. As Figure 5-2 shows, for the high negative
defocus region the aerial image center intensity predicted by simulation is higher than the
experimental values. This is not accounted for by the addition of astigmatism. Aerial image
simulations have shown that adding spherical aberration reduces the intensity for these focus
values. Figure 5-5 shows the effect of adding various levels of spherical aberration to the
simulation. There is a minor effect on the slope near zero defocus, a large lateral shift in the
curve, and a drop in the center intensity values for the negative defocus region. A spherical
aberration level of 0.1 waves matches the experimental results best in this region. The addition
of this spherical aberration to the optical model improved the center intensity match between
RADICAL simulation and the AIT images. Also, comparing Figure 5-3 and Figure 5-6 show
that the shape of the image is rounder and closer to the AIT image after the addition of the
spherical aberration. This suggests that spherical aberration is present and the match in the
center intensity is not a coincidence.
-10 -8 -6 -4 -2 0 2 4 6 8 100
0.5
1
1.5
2
2.5
Focus
Inte
nsity
Experiment: Defect 1
Experiment: Defect 2
Experiment: Defect 3
Simulation: astig=0.0
Simulation: astig=0.2
Simulation: astig=0.4
Simulation: astig=0.6
Simulation: astig=0.8
Intensity
Focus (µm)
Comparison of Intensity at Center of Defect
Figure 5-4. Change is center intensity through focus as a function of astigmatism. The astigmatism tends to
flatten the curve. The units of astigmatism are waves
44
5.3 Coma
When the astigmatism and spherical aberrations are included the center intensity of the
simulation matched the AIT experiments very well. But, comparing the images themselves
shows one subtle difference, an asymmetry in the intensity in the dark center area between the
northwest and southeast sides. This is due to a relatively small coma aberration. Figure 5-6
shows the comparison of an AIT image with RADICAL simulation for -4.6µm of defocus.
Three coma values are used as examples to show increasing asymmetry for higher values of
coma. A coma value of 0.1 matched the experimental images best.
The final plot of the center intensity comparison between the AIT images and RADICAL
simulations, which include all of the aberrations discussed above, is shown in Figure 5-7 along
with a few example images. These images show that the effect of the aberrations on the shapes
of the images is predicted correctly by RADICAL, even though the shapes of the images near
zero focus were not considered during aberration extraction.
Coma = 0 Coma =0.07 Coma = 0.14Experiment Simulation Simulation Simulation
Coma = 0 Coma =0.07 Coma = 0.14Experiment Simulation Simulation Simulation
Figure 5-6. Comparison of experimental aerial image with RADICAL simulation for negative defocus and
varying coma values. The asymmetry increases for higher coma values. For these images astigmatism =
0.55, and spherical = 0.10. The units for all aberrations are waves.
-12 -10 -8 -6 -4 -2 0 2 4 6 80.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Focus
Experiment: Defect 1
Experiment: Defect 2
Experiment: Defect 3
Simulation: Spherical = 0.00
Simulation: Spherical = 0.05
Simulation: Spherical = 0.1
Simulation: Spherical = 0.15
Simulation: Spherical = 0.2
Intensity
Focus (µm)
Comparison of Intensity at Center of Defect
Figure 5-5. Center intensity through focus as a function of spherical aberration. Spherical changes the best
focus position and lowers the center intensity value for very negative defocus values. The units of spherical
are waves.
45
5.4 Limitations of Method
The method described above showed a new way to determine aberrations given a limited
set of information about an optical system. But, its value is limited. In practice, varying
aberration values need to be determined across a large field. This means that a buried defect
would need to be imaged through focus at many points in the field of the tool. Then, the method
above would need to be repeated separately for each set of images. Also, only three aberrations
are extracted with this method. For the AIT, this was enough to match the experimental images.
This would likely not be true for an arbitrary system.
The final point to note is that the aberrations extracted in this work are much larger than
the aberrations expected in any EUV exposure tool. In 2009, for example, Nikon’s EUV optics
had an average wavefront error of 0.03 waves RMS [47]. The AIT had 0.55 waves RMS of
astigmatism alone. It is not clear that the method described above is sensitive enough for a
system with aberrations much lower than the AIT. It is not possible to test the sensitivity on the
AIT because of the signal to noise ratio is not good enough. Actinic imaging tools with a better
signal to noise ratio than the AIT do not currently exist.
5.5 Summary
This section showed that a series through focus images of a single buried defect with
known dimensions can be used to determine the aberrations of an inspection tool. In fact, most
of the aberrations could be determined by monitoring only one point, the center of the defect,
through focus. The great match between the simulation results and AIT images after the
aberrations are included shows again that RADICAL is accurate.
0 0.5 1
0
0.2
0.4
0.6
0.8
1
0 0.5 1
0
0.2
0.4
0.6
0.8
1
0 0.5 1
0
0.2
0.4
0.6
0.8
1
0 0.5 1
0
0.2
0.4
0.6
0.8
1
0 0.5 1
0
0.2
0.4
0.6
0.8
1
0 0.5 1
0
0.2
0.4
0.6
0.8
1
Focus = -3.6µm
Exp Sim Exp Sim Exp exp
Focus = 0.3µm Focus = 3.4µm
-12 -10 -8 -6 -4 -2 0 2 4 6 80.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Focus (um)
Inte
nsity
Comparison of Intensity at Center of Defect
Experiment: Defect 1
Experiment: Defect 2
Experiment: Defect 3
Simulation with Aberrations
Intensity
Focus (µm)
Comparison of Intensity at Center of Defect
Figure 5-7 Plot of center intensity of buried defect through focus comparison between RADICAL with
astigmatism = 0.55, spherical 0.10 and coma 0.06 waves and experimental AIT images.
46
6 Printability of Defects
Near Features The previous chapters show that isolated defects are useful to learn and demonstrate
general characteristics of buried defects in EUV masks. They also demonstrated the accuracy of
RADICAL. But, the real danger of buried defects in EUV masks is the effect they have on
features. This chapter studies, with simulation and experiment, the printability of buried defects
near features. The effects of defect size and position, as well as focus and illumination all have
an important effect on the printability of defects near features.
All simulations in this section will assume a production type EUV tool with NA = 0.32 and
4x reduction. The illumination is tophat with σ=0.75, except in the section on the effects of
illumination. The 13.5nm light is incident on the mask at a 6° angle perpendicular to the
direction of the line space patterns.
6.1 Effects of Defect Size
Chapter 4 showed the unsurprising result that the taller isolated defects are, the more
printable they are. This is true of defects near features as well. Figure 6-1 shows the space CD
of 22nm dense lines in the presence of a Gaussian defect 25 nm from the shadowed absorber
edge on the 4x mask. No smoothing was applied, so both the surface defect width and height
could be controlled.
Figure 6-1. Summary of space CD for 22nm dense lines as a function of defect size and shape.
One interesting result shown in Figure 6-1 is that printability does not always get worse for
wider defects. For narrow defects the behavior is not surprising. A very narrow defect is not
20 40 60 80 100
0.5
1
1.5
2
2.5
3
3.5
4
Surface Defect FWHM (nm)
Surface D
efe
ct H
eig
ht (n
m)
0
5
10
15
20
47
printable and as the defect FWHM increases so does the resulting CD change. But, this is only
true until the FWHM reaches values around 65nm. At this point, the CD change due to the
defect begins to decrease as the FWHM increases. This effect was also shown by Terasawa in
[48].
6.2 Effects of Defect Position
As described in Chapter 3, RADICAL’s design makes simulating several different
absorber patterns over the same defect fast and easy. Therefore, the investigations in this chapter
will simulate the effect of a defect as a function of position across an entire period on the mask.
This is not usually done in simulation or experimental studies. The resulting CD change as a
function of position for several defects is shown in Figure 6-2. The offset is in 4x mask
dimensions and the CD is in 1x wafer dimensions.
Figure 6-2. Space CD of 22nm dense lines as a function of defect position for multiple defect sizes and tophat
illumination with a partial coherence of σ=0.75.
Several interesting conclusions can be drawn from these results. The first is that a 2.2nm
tall defect will cause resist bridging if it is located anywhere between 55nm and 75nm from the
absorber edge. This is the worst case area in which a defect can be located. However, in
practice many defects that do not cause feature bridging will still not be acceptable because the
tolerable CD change due to a defect is actually a small percentage of the space width. For this
work, 10% will be used as the maximum allowable space CD change. The RADICAL
simulations in Figure 6-2 show that a defect as small as 0.8nm will cause a 10% CD change if it
is located 65nm from the edge of the absorber line. This means if a mask blank has no defects
taller than 0.8nm it can be relied on for EUV printing of 22nm dense lines, assuming no focus
variation.
Figure 6-2 is not symmetric due to the 6° incident angle of the EUV light. For negative
offsets, the size of the space is actually increased by the defect and for positive offsets the space
width is reduced. Also, all of the curves in Figure 6-2 are shifted right due to the 6° incident
angle. For normally incident light, the curves would be symmetric around an offset of +44nm,
where the defect is centered in the space. But, in Figure 6-2 the curves are roughly symmetric
around an offset of +66nm, 22nm different than the expectation for normal incidence. The
-40 -20 0 20 40 60 80 100 1200
5
10
15
20
25
Defect Offset (nm)
CD
(nm
)
Space CD as a Function of Defect Position
Surface Defect Height: 0.8nm
Surface Defect Height: 1.4nm
Surface Defect Height: 2.2nm
Surface Defect Height: 2.8nm
Nominal CD
10% CD Change
48
absorber stack in these simulations was 87nm tall, 75nm of TaN and 12nm of an anti-reflection
coating for mask inspection. A simple geometric ray tracing calculation of the round trip
shadowing distance predicts a horizontal displacement of 18nm from the normal incidence case.
This is close to the value of 22nm observed in simulation, suggesting geometric approximations
are a good first pass method for predicting shadowing effects. But, the most important
conclusions regarding shadowing is that when considering the effect of a defect, its position and
the angle of the incident light must be considered together.
Figure 6-3 shows the results for 16nm dense lines. All of the characteristics of the 22nm
results are present in the 16nm results, but the sensitivity to defects is increased greatly. This is
due mainly to the reduced image slope from 22nm to 16nm patterns. For 16nm lines and this
system, k1=0.38. It is likely that in production resolution enhancement techniques such as off-
axis illumination would be used to improve the image slope. This will be addresses in Section
6.4.
Figure 6-3. Critical dimension of 16nm dense lines as a function of defect position for multiple defect sizes and
tophat illumination with a partial coherence of σ=0.75.
6.3 Effects of Focus
The results in Chapter 4 show the clear effect of focus on buried EUV defects. Both
RADICAL simulations and AIT images of isolated buried defects invert their center intensity
through focus. The center intensity of the aerial image of an isolated bump defect is below the
clear field value for negative focus, which is defined as moving the wafer away from the lens,
and above the clear field value for positive focus. This inversion through focus is apparent in the
images of buried defects near features as well. Figure 6-4 shows the resulting space CD for
nominally 22nm lines as a function of the position of 0.8nm tall defect for three focus values.
-20 0 20 40 60 800
2
4
6
8
10
12
14
16
18
20
Defect Offset (nm)
CD
(nm
)
Space CD as a Function of Defect Position for 16nm Dense Lines
Surface Defect Height: 0.4nm
Surface Defect Height: 0.8nm
Surface Defect Height: 1.4nm
Surface Defect Height: 2.2nm
Nominal CD
10% CD Change
49
Figure 6-4. CD for 22nm lines as a function of the position of 0.8nm tall defect for three focus values
Figure 6-4 shows that for negative focus the space CD is decreased and for positive focus
the space CD is increased. This is consistent with Chapter 4, where the defect is brighter for
positive focus and darker for negative focus. There are a few interesting offset locations to note.
From 15nm to 50nm offset, the CD change is within the 10% boundary at best focus, but out of
focus it is beyond the limit. This shows that focus must be considered when minimum defect
sizes are defined. The other interesting position is at 10nm offset, where there is no CD change
due to the defect in focus, but out of focus the CD change is nearly 10%. This shows that
inspection through focus may be necessary for defect detection.
6.4 Effects of Illumination
The primary reason advanced illuminations, such as annular and dipole, are used is to
improve the quality of an aerial image. A standard measure of image quality is contrast. Figure
6-5 shows the improvement of image contrast of 16nm and 22nm dense lines for three
illuminations.
Figure 6-5. Contrast of the aerial images of 22nm and 16nm dense line patterns with three illuminations.
-40 -20 0 20 40 60 80 100 12015
16
17
18
19
20
21
22
23
24
25
Defect Offset (nm)
CD
(nm
)
Space CD as a Function of Defect Position
defocus: -75nm
defocus: 0nm
defocus: 75nm
Nominal CD
10% CD Change
0 0.2 0.4 0.6 0.8 1
TopHat
Annular
Dipole
16nm
22nm
Contrast
50
Section 4.3 showed that illumination has a large effect on the images of an isolated defect.
It also has a large effect on the printability of defect near features, as shown in Figure 6-6 which
shows a reduction of defect printability for annular and dipole illuminations at best focus. The
reduction in sensitivity seems to follow the same trend as the increase in contrast suggesting that
a way to decrease defect printability is to increase contrast.
Through focus, however, the off-axis illuminations do not always reduce defect
printability. As Figure 6-7 shows, the defect has a more complicated effect for annular and
dipole illuminations through focus than it does for tophat. The space containing the defect with
tophat simply expands and contracts near the defect through focus, in the same way it did for
previous simulations in this chapter. But, the annular and dipole illuminations cause the defect
to affect a larger area of the image. The dipole image is particularly interesting. In focus, the
image appears to be defect free. But, for positive defocus the space CD is decreased above and
below the defect and increase to the left and right of the defect. This is consistent with the
isolated image of the defect in Figure 4-16. The effect is opposite for negative focus.
For annular illumination, similar effects over a large area are observed, but the magnitude
of the changes due to the defect is much smaller. This suggests that annular illumination may be
the best illumination for printing 16nm dense lines that are insensitive to defects through focus.
The mutual intensity function for each illumination is also shown in Figure 6-7. The dark areas
of the mutual intensity function show the points where there is no interaction with the fields from
the center point. The orange areas show positive interaction and the blue areas show negative
interaction. The dipole mutual intensity function shows that the defect interacts strongly over the
entire domain, so the effects observed in the aerial image plots are not surprising.
Figure 6-6. CD for 16nm lines as a function of the position of 0.8nm tall defect for three types of illumination.
-20 0 20 40 60 8011
12
13
14
15
16
17
18
19
Defect Offset (nm)
CD
(nm
)
Space CD as a function of illumination for a 0.8nm surface defect and 16nm dense lines
TopHat
Annular
Dipole
Nominal CD
10% CD Change
51
Figure 6-7. Aerial images through focus for three illuminations of a 0.8nm defect 16nm (mask scale) the edge of an
absorber line in a 16nm (wafer scale) dense line space pattern. The mutual intensity function for each illumination
is also shown. The physical size for all plots is 128nm x 128nm wafer scale.
6.5 Experimental Images of Defects Near Features
As was done for isolated defects, actinic inspection images of the Intel programmed defect
mask from the AIT can be compared directly to RADICAL simulation results. The sizes of all
the defect images are summarized in Table 2-1. All of the images discussed in this section will
be for defects located near 250nm dense lines. All dimensions are mask scale. Three nominally
identical defects were programmed at the same relative location within a line space pattern.
6.5.1 Printability as a Function of Defect Size
An example AIT image is shown in Figure 6-8 showing three defect sizes with three
defects each, located 55nm away from the edge of an absorber line.
f = -75nm f = 0nm f = +75nm
Tophat
Annular
Dipole
-60 -40 -20 0 20 40 60
-60
-40
-20
0
20
40
60
-60 -40 -20 0 20 40 60
-60
-40
-20
0
20
40
60
Mutual Intensity
52
Figure 6-8. AIT image at focus = -0.75µm of nine programmed defects located 55nm away from the edge of 250nm
absorber lines. There are three defects of each size.
A summary of the CD change due to these defects is plotted in Figure 6-9 along with the
CD change predicted by RADICAL simulations. All of this analysis was done for images -
0.75µm out of focus. These happened to be the best quality AIT images, so they were used.
There are three sets of experimental data, one for each column on the programmed defects mask.
Several conclusions can be drawn from Figure 6-9. The first is that RADICAL matches the AIT
images fairly well up to the 6.5nm tall defect. This is actually a very large defect, larger than the
defects used in the accuracy studies presented in Chapter 3. The defects which are important for
the 22nm node and below will only one-third the size at about 2nm tall and bel0w. Another
interesting feature of Figure 6-9 is that the CD change due to defects in column 1 is nearly
always less than the CD change due to defects in column 3. This is because the focal position of
the AIT image is not uniform. Due to details of the zone plate imaging system, which are
beyond the scope of this dissertation, there is a gradual focus change across the image. Because
buried defects are so sensitive to focus, there is a systematic variation in CD across the image.
X Distance (um)
Y D
ista
nce (um
)
14 15 16 17 18 19
13
14
15
16
17
18
19
20
Defect
Column 3
3.5nm
Defects
Defect
Column 2
2.7nm
Defects
2.0nm
Defects
Defect
Column 1
53
Figure 6-9. CD change as a function of surface defect height at focus = -0.75µm from AIT experiments and
predicted by RADICAL simulations. A positive CD change in this plot is defined as a decrease in the space CD.
The final interesting result in Figure 6-9 is the 4.4nm tall defect in column 2 that appears to
cause the space CD to increase. Looking at the images of this defect in Figure 6-10 shows that
the CD increase is present at multiple focus positions and only in column 2. This means that it is
likely an additional non-programmed defect on the mask. This is possibly due to some absorber
missing near the defect.
-100
-50
0
50
100
150
200
250
0 1 2 3 4 5 6 7 8
CD
Ch
an
ge
(n
m)
Surface Defect Height (nm)
RADICAL Simulation
AIT Defect Column 1
AIT Defect Column 2
AIT Defect Column 3
54
Figure 6-10. Images near best focus for the three nominally 4.4nm tall defects.
It may be tempting to use the data in Figure 6-10 and to determine the allowable defect
sizes for production EUV lithography. However, the differences in dimensions and illumination
between these experiments and future production conditions mean that final quantitative
conclusions cannot be reached from this data. The sizes of the features are more than two times
larger than the patterns that will be printed with production EUV. Also, the illumination of the
AIT is very coherent and on-axis. Production EUV will either use a less coherent tophat
illumination or off-axis illuminations like dipole or annular, as demonstrated by simulations
earlier in this chapter. Still, several qualitative conclusions can be drawn from the experimental
data.
6.5.2 Printability of Covered Defects
Some have proposed that a solution to blank defects is to shift the absorber pattern to cover
them [49]. This may be a plausible solution which makes a few blank defects tolerable, but it
must be done carefully. Figure 6-11 shows the resulting simulated and experimental images of a
7nm tall 59nm FWHM defect that is centered 43nm from the edge under the 250nm absorber
line. Figure 6-11 shows there is a small bump in the AIT image near the defect. The CD drops
by as much as 23nm here. The results of the simulation show a similar bump, with a 19nm CD
change. For 250 lines, these changes are less than 10%. But, for 88nm lines on the mask, which
would print 22nm wafer lines in a 4x system, 19nm would be an over 20% change. These
experimental and simulated results both show that a covered defect may print, so it is important
to fully consider all defects, even if they are covered by absorber features.
0.5 1 1.5 2
0.5
1
1.50.5 1 1.5 2
0.5
1
1.50.5 1 1.5 2
0.5
1
1.5
0.5 1 1.5 2
0.5
1
1.50.5 1 1.5 2
0.5
1
1.50.5 1 1.5 2
0.5
1
1.5
0.5 1 1.5 2
0.5
1
1.50.5 1 1.5 2
0.5
1
1.50.5 1 1.5 2
0.5
1
1.5
f = 0.75µm f = 0µm f = -0.75µmC
olu
mn
1C
olu
mn
2C
olu
mn
3
55
In this case, the defect is 7nm tall at its center, and has a full width half max of 59nm.
According to the growth model in [43], which predicts an approximately Gaussian surface
profile, the height of the defect at the edge of the absorber pattern is 2.2nm, which corresponds to
a round trip path difference for incident light of 0.3λ. This significant fraction of a wavelength
should interfere and cause a distortion in the image, therefore the results in Figure 6-11 are not
surprising. So, although the defect appears to be covered by the absorber it still causes
meaningful mask geometry in the space between the lines.
Figure 6-11. AIT Image, Mask Layout and RADICAL image for covered defect.
6.5.3 Printability Through Focus
It has been well established in this dissertation that buried defects are phase defects and
their printability is very sensitive to focus. This is shown by AIT images and RADICAL
simulation for a defect near features in Figure 6-12. Qualitatively the two rows of images are
very similar which suggests that RADICAL can accurately predict the experimental printability
of buried defect near features through focus. Quantitative comparisons of the CD change,
however, were very difficult because of the high source coherence. The source coherence causes
a relatively narrow bright spot near the defect for positive focus. This aspect, present in the AIT
images and RADICAL simulations, made it unrealistic to quantitatively compare the CD change
due to the defect through focus
AIT Image Actual Mask Layout RADICAL Image
56
Figure 6-12. Comparison of aerial images of 5.3nm buried defect on the edge of an absorber line predicted by
RADICAL and recorded on the AIT. The label on each image is focus.
6.6 Summary
This chapter presented a thorough study of defect printability near features. The
printability of defects is very sensitive to their position relative to absorber features. The worst
case position is a function of the defect size, and even defects covered by the absorber pattern
can cause a CD change greater than 10%. The sensitivity to defects was decreased in focus by
advanced illuminations. But, through focus the defects affected a larger area of the aerial image
with dipole and annular illuminations compared to tophat. Many of the effects observed in this
chapter were verified experimentally by comparisons between RADICAL simulations and AIT
images, confirming again RADICAL’s accuracy.
-1.0μm -0.25μm 0.5μm 2μm0 8
2 2.2 2.4 2.6 2.8 3 3.2
2 2.2 2.4 2.6 2.8 3 3.2
2 2.2 2.4 2.6 2.8 3 3.2
0 8
2 2.2 2.4 2.6 2.8 3 3.2
0.2
0.4
0.6
0.8
1
1.2
1.4
1.25μm
AIT Images
RADICAL Images (Simulation)-1.0μm -0.25μm 2μm1.25μm
(Experiment - December 2008)
57
7 Compensation Methods
for Buried Defects This chapter addresses mask infrastructure needs by investigating, with simulation,
compensation methods for EUV masks with buried defects. All simulations in this work are
performed with RADICAL, which was described in detail in Chapter 3. Compensation has been
investigated previously with simulation [9] and experiments [50]. But, neither of these works
addressed the effectiveness of the compensation through focus, which this dissertation shows is a
challenging problem.
In Chapter 2, many of the issues facing EUV lithography were described. In this chapter, a
possible solution to the primary issue, masks defects, is explored. Currently, the density of
defects detected on EUV masks is 10 times higher than required defect densities [22]. It is
possible that all other components of the EUV system may be in place, but EUV masks will still
not be available which meet the required defect density.
The general goal of compensation is to take an EUV mask with an unacceptable yield and
transform it into a mask with an acceptable yield. The specifics of this goal will vary for each
design and process, but for this work it is assumed that a CD change greater than 10% is not
acceptable. It turns out that for the defects investigated in this chapter reducing the CD to less
than 10% is not difficult in focus. But, due to the phase nature of the buried defect the CD
change is much worse out of focus. Therefore, compensation methods must not only reduce the
CD at best focus to less than 10% but also reduce the CD variation due to the defect through
focus.
Two methods will be presented in this chapter to compensate for buried defects in EUV
masks. The first is a design curve method. For this method, design curves developed from the
results of many defect free simulations will be used to determine the required absorber
modification based only on the CD change due to the defect. This method works well in focus,
but the defect still has a significant effect through focus. The second method attempts to
compensate for a buried defect through focus by covering the defect with absorber. This method
requires knowledge of the defect’s size and position, but is able to reduce the defect’s effect
through focus. The effectiveness of these compensation methods will be demonstrated on
example defects.
All simulations for this chapter are for 22nm dense lines (wafer scale), 4x demagnification,
NA=0.32 and a tophat illumination with σ=0.75 incident at 6°. These parameters are chosen to
model a first generation production EUV lithography scanner. Also, positive focus refers to
moving the wafer closer to the optics and negative focus refers to moving it away from the
optics. The defect size and position values are mask scale. The CD will be measured at the
wafer scale and, because buried defects normally affect two adjacent dark lines, the term CD in
this chapter refers to the space CD, as shown in Figure 7-4.
7.1 Compensation with Design Curves
Compensation with design curves, the first method proposed in this paper, is a simple and
general method suitable for using inspection or wafer print data to guide compensation. The
58
assumption is the CD change due to a defect is known. Given this CD change the design curves
will then prescribe several possible modifications to the absorber pattern to compensate for the
defect. It is important to note that no more information is required than the CD change. The
size, shape and position of the defect are not required.
The fundamental assumption of the design curve method is that a geometry that causes a
certain percentage CD increase for a defect free simulation will compensate for a defect that
causes that same percentage CD decrease before compensation. This assumption is not exactly
accurate. The interaction of the electric fields at the mask and partial coherence effects of the
illumination and imaging system are much more complex than that. But, this simple method
does work fairly well for many defects.
7.1.1 Development of Design Curves
There are many complicated geometries
that could be used for compensation. But, for
manufacturability and to reduce the number of
variables in this study, only symmetric
geometries with rectangular sections of absorber
removed are used for the design curve method.
Therefore, the compensation geometry is
specified by two variables: the length and the
depth of the removed absorber. These parameters
are defined on the example geometry shown in
Figure 7-1.
Simulations were performed for a range of
length and depth values. The results are
summarized in Figure 7-2. It is interesting to
note that the percentage CD change is
approximately a function of the removed area.
This is a useful fact to remember when considering compensation geometries, because one
dimension can be reduced and the other increased, as long as the area is held constant, to
improve manufacturability.
Example design curves derived from the data above are shown in Figure 7-3. Converting
the data in Figure 7-2 into the plots in Figure 7-3 was a three step process. The first step was to
fit the data for each length with a third degree polynomial to produce five expressions, one for
each length, for depth as a function of percent CD change. The second step was to use these
expressions to determine the depth of compensation needed for each length, for the percent CD
changes shown in Figure 7-3. This produced sets of depth and length pairs for each CD change.
Finally, each one of these sets was fit with a third degree polynomial to produce the plots below.
A specific curve can be calculated quickly for any CD change within the range of CD changes
produced by the defect free simulation.
Absorber Pattern on Mask
X Distance (nm)
Y D
ista
nce (nm
)
0 200 400 600
0
100
200
300
400
500
600
700
D
L
Figure 7-1. Example of compensated geometry with
definitions of depth (D) and length (L)
59
Figure 7-3. Design curves for specifying compensation geometries for defects causes 10%, 15%, 20% and 25% CD
changes
0 5 10 15 20 2550
60
70
80
90
100
110
Depth of Removed Area (nm)
Length
of R
em
oved A
rea (nm
)
Dimensions of Removed Area for Several CD Changes
∆CD = 10%
∆CD = 15%
∆CD = 20%
∆CD = 25%
Distance (nm)
Dis
tance (nm
)
Defective Image Before Compensation
50 100 150
20
40
60
80
100
120
140
160
CD
Figure 7-4. Aerial image at best focus from a 0.8nm tall 108nm FWHM defect located 30nm from the
shadowed edge of the absorber pattern.
0 5 10 150
5
10
15
20
25
30
Depth of Removed Area (nm)
% C
D C
hange
Percent CD Change as a Function of Absorber Removed
Length: 50nm
Length: 65nm
Length: 80nm
Length: 95nm
Length: 110nm
0 500 1000 1500 2000 2500 3000 35000
5
10
15
20
25
30
% C
D C
hange
Total Area of Removed Absorber (nm2)
Percent CD Change as a Function of Removed Area
Figure 7-2. Summary of in focus defect free simulations used to develop design curves
60
7.1.2 Example of Compensation with Design Curves
Design curve compensation will be demonstrated for a defect that is 0.8nm tall, has a full
width at half maximum (FWHM) of 108nm and is located 30nm from the shadowed edge of the
88nm dense line space pattern. All these dimensions are mask scale, which means the 88nm
lines will be printed as 22nm lines due to the demagnification. This defect causes a 15% CD
change at best focus, as shown in Figure 7-4. The design curve in Figure 7-3 for a 15% CD
change specifies the possible geometries to account for this defect. Three selected compensation
geometries are summarized in
Table 7-1. The resulting aerial images through focus of two of the geometries are shown
along with the images from the uncompensated geometry in Figure 7-5. The space CD measured
for each image is plotted in Figure 7-6.
Table 7-1.Summary of compensation geometries for a 0.8nm x 108nm defect
Geometry Label Length Depth
A 56nm 10nm
B 72nm 8nm
C 108nm 6nm
Figure 7-5. Through focus images of the original uncompensated geometry and compensation geometries A and C.
A
C
f = -80nm f = -40nm f = +40nm f = +80nmf = 0nm
Ori
gin
al
(No
Co
mp
en
sati
on
)
61
Figure 7-6. Plot of percent CD change for the three compensation geometries plus the original geometry through
focus.
Figure 7-5 and Figure 7-6 show that the design curve method for compensation is
successful in focus. The CD change is improved from a decrease of 15% to an increase of 3%.
This is well within the 10% CD change requirement. The compensation works so well in focus,
that there is no defect visible in the center column of the compensated images. However, the
defect does become visible through focus. Buried defects are phase defects so unlike amplitude
defects, the effect of phase defects inverts through focus. This causes the contraction on the lines
for negative focus in Figure 7-5 and the expansion of the lines for positive focus. The change is
represented by the slope in Figure 7-6. The slope before and after compensation is
approximately the same. This means that the design curve method is able to correct for the
defect in focus but it is not able to correct for it through focus. Improving the through focus
compensation for the defect is addressed in the next section.
7.2 Compensation by Defect Covering
As shown above, buried defects are phase defects and therefore print worse out of focus.
One possible method to counteract this effect is to cover a defect with absorber. The goal is to
transform a phase defect into an amplitude defect. Amplitude defects print worst in focus, so if
they can be compensated for in focus the effect of the defect will be acceptable through focus.
7.2.1 Example of Compensation by Defect Covering
The defect covering strategy can be applied to the defect from Section 7.1. Unlike the
design curve method, the defect covering method does require that the location, size and shape of
the defect is known because full simulations of the defect are required. Rather than simply
changing the absorber based the results of previous simulations, as is done in the design curve
method, many in focus simulations will be performed for the actual defect that is being
compensated. The results of this compensation are shown in Figure 7-7 and Figure 7-8.
A 26nm x 26nm absorber square was chosen to cover the defect. Defects larger than this
size caused the intensity of the line near the square to be reduced to a level that was assumed to
be too low to effectively expose resist on a wafer.
-80 -60 -40 -20 0 20 40 60 80-50
-40
-30
-20
-10
0
10
20
Focus (nm)
% C
D C
hange
Percent CD Change Through Focus
Compensation A
Compensation B
Compensation C
No Defect or Compensation
Defect without Compensation
10% CD Change Limit
62
Table 7-2. Summary of Geometries in Figure 7-7
Square Size Length Depth (Left) Depth (Right)
Design Curve Method none 72nm 8nm 8nm
Covered Defect with Balanced Absorber Removal 26nm x 26nm 80nm 18nm 18nm
Covered Defect with Unbalanced Absorber Removal 26nm x 26nm 80nm 24nm 12nm
Figure 7-7. Compensated layouts along with in focus aerial image for three compensation schemes for a 0.8x108nm
defect.
Figure 7-8. CD change for three compensation schemes through focus for a 0.8x108nm defect
The first attempt at covering the defect removed absorber from the lines near the defect
symmetrically as was done in the design curve method. The exact compensation applied is
Design Curve
MethodCovered Defect
Covered Defect
w/Unbalanced
Removal
Pa
tte
rnB
est
Fo
cus
Ima
ge
-80 -60 -40 -20 0 20 40 60 80-25
-20
-15
-10
-5
0
5
10
15
20
Focus (nm)
% C
D C
hange
Percent CD Change Through Focus
Compensation WITHOUT Covering (Design Curve)
Compensation WITH Covering
Compensation WITH Covering and Unbalanced Removal
No Defect
10% CD Change Limit
63
summarized in Table 7-2. This resulted in a CD that met the met the specification in focus. It
also increased the depth of focus (DOF), defined in this case as the focus range for which the
absolute value of the CD change remained below 10%, by 13% over the design curve method
from 112nm to 126nm. The amount of absorber removed from each was greater than for the
design curve method because the absorber covering the defect blocked light that needed to be
replaced. Unfortunately, while this compensation geometry produced a moderate improvement
in depth of focus, it also causes some bending in the image of the line. To correct for the
bending of the image, a different amount of absorber needed to be removed from each side. But,
the total area removed remained the same. This unbalanced compensation geometry produced an
in focus image with the correct CD and no bending, but it reduced the depth of focus compared
to the compensation with covering and balanced absorber removal. The depth of focus is 118nm
which is only 5% better than the depth of focus of the design curve method geometry.
The fact that none of these covering techniques were able to significantly improve the
through focus behavior shows that the attempt to transform a phase defect into an amplitude
defect was not successful. This is understandable, because the defect has a FWHM of over
100nm, but the squares used to cover it are less than 25nm on each side. Therefore, only a small
portion of the defect is actually covered. This logic also explains why the depth of focus was
worse for the unbalanced absorber removal. Removing more of the absorber near the defect
causes parts of the defect that were previously covered to reflect light that is slightly out of phase
with the background and contributed to the CD change through focus. If this analysis is correct,
then a smaller defect should be compensated for more effectively through focus by covering.
7.2.2 Example of Compensation by Defect Covering for a Narrower Defect
To test whether a narrower defect can be covered more effectively, a second defect is
compensated for by the design curve method and defect covering method. This defect is 1.6nm
tall and 50nm FWHM on the surface of the mask blank. It is located in the center of the space
between two absorber lines. This defect causes a CD change of 17% at best focus without
compensation. The results of the compensation are shown in Figure 7-9.
Figure 7-9. CD change for three compensation schemes through focus for a 1.6x50nm defect
-80 -60 -40 -20 0 20 40 60 80-60
-50
-40
-30
-20
-10
0
10
20
Focus (nm)
% C
D C
hange
CD Change for Three Absorber Geometries
Defect without Compensation
Compensation by Removing Absorber (Design Curve Method)
Compensation by Covering Defect and Removing Absorber
Nominal CD
10% CD Change Limit
64
Table 7-3. Summary of geometries used to compensate for 1.6x50nm defect
Square Size Length Depth (Left) Depth (Right)
Design Curve Method none 84nm 8nm 8nm
Covered Defect 26nm x 26nm 65nm 18nm 18nm
For this narrower defect, the depth of focus was increased from 100nm, for the design
curve method, to 123nm for the covering method. This is a 23% increase. This improvement of
DOF is much better than for the larger defect, which shows that the effectiveness of the covering
method is related to the percentage of the defect area that is covered. Line bending was not an
issue for this defect. Although there was improvement in the depth of focus, the effects of the
defect’s phase are still obvious through focus, suggesting it may be impossible to transform even
this smaller defect into a magnitude defect.
7.3 Compensation of Buried Pits
For some EUV mask blanks, pits are actually more common than bumps [51]. Therefore,
it is important to show that these compensation methods work for pits as well. The pit to be
compensated for is 2nm deep, 50nm FWHM, and located in the center of the space between two
absorber lines. This defect, like the 50nm wide bump, caused a 17% CD change. The results of
the compensation are shown in Figure 7-10. It is important to note that like the bumps shown
above, pits cause a reduction in the space CD. So although they may seem to be the opposite of
bumps, the same design curves which specify absorber removal can be used.
Figure 7-10 demonstrates a few interesting things. The first is that the design curve
method works for pits in the same way it works for bumps. The major difference between the
printability of pits and bumps is also obvious in the decrease in space CD from negative to
positive focus. As Figure 7-6, Figure 7-8, and Figure 7-9 show the space CD near bumps
increases from negative to positive focus. The depth of focus is increased 23% from 83nm for
the design curve compensation to 102nm for the covering method. The covering method works
as well for pits as it does for bumps.
Figure 7-10. CD change for three compensation schemes through focus for a -2.0x50nm defect
-80 -60 -40 -20 0 20 40 60 80-80
-60
-40
-20
0
20
Focus (nm)
% C
D C
hange
CD Change for Three Absorber Geometries
Defect without Compensation
Removing Absorber (Design Curve Method)
Covering Defect and Removing Absorber
65
Table 7-4. Summary of geometries used to compensate for -2.0x50nm defect
Square Size Length Depth (Left) Depth (Right)
Design Curve Method none 70nm 10nm 10nm
Covered Defect 26nm x 26nm 90nm 30nm 14nm
7.4 Summary
Printability of buried defects in EUV masks can be decreased by adding or removing
absorber near the defect. This may allow a defective mask to be transformed into a useable mask
with simple modifications to the absorber pattern near the defects. Design curves, developed by
defect free simulations of possible compensation geometries, can accurately prescribe absorber
modifications for a defect based on the CD change due to that defect alone. Unfortunately, the
phase of the buried defects still causes a through focus CD variation, even after the compensation
is applied. For a 1.6nm tall and 50nm wide defect, the DOF for the design curve method was
only 100nm. The method proposed to account for this was to cover the defect with absorber and
then increase the amount of absorber removed to produce the correct CD in focus. This
improved the DOF to 123nm for the 1.6nm tall defect, though the phase effects of the defect still
had a major effect on the CD through focus. These two methods work equally well for pits. The
results of this chapter can also guide technology development. Specifically, because narrow
defects can be more easily compensated through focus, smoothing processes should be
developed which produce narrower surface defects.
66
8 Conclusion This dissertation presented a new simulator specifically designed for EUV masks with
buried defects, RADICAL, and demonstrated several applications of the software that have
already contributed significantly to EUV research and development. This simulator is an
extension of the multilayer blank simulator developed by Michael Lam. The primary advantage
of RADICAL over other simulation methods is its speed. It can simulate a geometry in a minute
or less that would take hours using conventional rigorous methods. This speed increase was
accomplished by evaluating the electromagnetic response of the two components of the mask,
the absorber and non-planar multilayer, and the interactions between then, and specifically
designing a modular simulator to model these components and simulate their interaction as fast
as possible.
Within RADICAL, the absorber is modeled using a propagated thin mask model and the
multilayer is simulated by an advanced ray tracing method. The two components are linked by a
Fourier transform which converts the nearfield output of one simulator to a set of plane waves to
be input into the other. Multiple reflections between the absorber and multilayer do not need to
be considered because the similar refractive index values of all the mask materials for 13.5nm
light make the reflection coefficient off the bottom of the absorber insignificant. The single
surface approximation was tuned to account for the penetration of light into the multilayer and
can be used in place of the ray tracing simulator for shorter defects with minimal smoothing,
which increases the speed of RADICAL by two orders of magnitude. The accuracy was tested
by comparisons to two rigorous methods, the finite difference time domain and waveguide, and
actinic inspection experiments on the Actinic Inspection Tool (AIT) at Lawrence Berkeley
National Lab.
RADICAL has been used to learn many new things about buried defects, their interactions
with features, and the inspection tools that will be needed to inspect for them on EUV masks.
Studies of isolated defects showed that the dominant characteristic of buried defects is the phase
difference between the defect and the background light. This causes the inversion of the aerial
image intensity through focus, which has been observed by many researchers in multiple
simulations and experiments. But, simply assuming the defect is small and produces a constant
phase change will not yield the correct results. Due to the multilayer smoothing processes used
by mask blank suppliers, the surface geometries produced by buried defects are wide compared
to the wavelength of light and their reflected spectrum does not fill the pupil. Therefore, EUV
buried defects behave differently than the defects found on conventional DUV lithography masks
and are more difficult to model and compensate for.
This dissertation presented a thorough study of buried defects near features with a
particular focus on the effect of defect position on critical dimension (CD) change. Previous
time consuming rigorous simulations and expensive and inexact experiments were unable to do
careful studies of defect printability as a function of position. RADICAL, however, is able to re-
use the results of the multilayer simulations of a buried defect for any pattern near that defect,
making simulations of many defect positions very fast. These investigations showed that buried
defects are very sensitive to position, and the worst case position of a defect is actually a function
of the defect’s size. Also, as expected from the results of isolated defect imaging investigations,
the printability of buried defects is worst out of focus. Some defects may be invisible to an in
focus inspection, but out of focus can cause an unacceptable CD change. Examples were also
67
shown of defects that were covered by the absorber still causing a meaningful CD change. The
printability of covered defects was confirmed by an AIT inspection of an EUV programmed
defect mask, designed and fabricated by Intel, and shows that attempts to compensate for defects
by shifting the absorber pattern to cover them must be done carefully. RADICAL was used to
explore possible future production conditions with advanced illuminations, such as annular or
dipole. These illuminations improved image slope and therefore decreased the sensitivity to
buried defect in focus. But, through focus buried defects caused pattern distortions over a much
larger area on the wafer when advanced illuminations were used.
After presenting many examples of how problematic buried defects are for EUV printing,
solutions were offered. Compensation methods were described that may allow the use of mask
blanks with buried defects to create defect free masks. A design curve method was presented
than can compensate for a buried pit or bump in focus based simply on its CD change, as
measured by mask inspection or on a printed wafer. Another method was presented which can
improve the compensation through focus, but it requires detailed knowledge of the size, shape
and position of the defect so that it can be simulated in RADICAL.
The results of applications presented in this dissertation are for specific defective
multilayer geometries, mask patterns, and materials. It remains to be seen whether or not these
will be the final choices for EUV production, or what other new technologies will be proposed.
Whatever happens, RADICAL will continue to be a fast and accurate option for EUV mask
simulation. As new defects are discovered, RADICAL will help uncover their physical nature.
As new smoothing, inspection or compensation schemes are introduced, RADICAL will be able
to evaluate their effectiveness. And as EUV technology continues to be refined, RADICAL will
be able to help engineers quickly evaluate trade-offs between competing figures of merit.
The methods in this dissertation will also be useful once EUV lithography is implemented.
Masks will likely never be completely free of buried defects, and for each new design RADICAL
will be able to determine the tolerances of mask patterns to buried defects. In inspection
systems, fast defect simulation will allow real-time disposition of the defects detected.
RADICAL can also be integrated into OPC algorithms to allow automated pattern changes to
compensate for buried defects.
EUV lithography masks are not the only structures in which objects are buried under
multiple layers, and therefore the methods in this work could be applied to other fields. The
layers which make up the earth’s crust could be viewed in the same way as an EUV mask, and
the signature of oil or minerals below the surface may be similar to buried defects. The walls of
a building are also composed of many layers, so the methods in this paper may assist efforts to
image objects through walls. There are many problems that the methods in this dissertation
could be applied to, and time will tell what new issues within the semiconductor industry and
beyond find RADICAL and its technology valuable.
68
9 References
[1] D. Medeiros, et al. “Lithography on the Edge” presented at 2009 Int. SEMATECH EUVL
Symp., Prague (October 2009).
[2] S. Sivakumar, et al. “Technical and Manufacturing Challenges and the Prospects for HVM
using ArF Pitch Division” presented at 2009 Int. SEMATECH EUVL Symp., Prague
(October 2009).
[3] Wood, O. et al., “EUV lithography at the 22nm technology node,” Proc. SPIE 7636,
(2010).
[4] P. Kuerz, et al. “Optics for EUV Lithography” presented at 2009 Int. SEMATECH EUVL
Symp., Prague (October 2009).
[5] Miura, T. et al. “Nikon EUVL development progress update,” Proc. SPIE 7636, (2010).
[6] D. C. Brandt, et al. “LPP Source System Development for HVM,” SPIE 7636, (2010).
[7] B. LaFontaine, et al. “Assessing EUV Mask Defectivity” presented at 2009 Int.
SEMATECH EUVL Symp., Prague (21 Oct. 2009).
[8] H.S Seo1, et al. “Effects of multilayer depositions on the EUV printability and DUV
inspectability of substrate pit defects” presented at 2009 Int. SEMATECH EUVL Symp.,
Prague (October 2009).
[9] R. Jonckheere, et al. “Investigation of Mask Defect Density in Full-Field EUV
Lithography,” presented at 2009 Int. SEMATECH EUVL Symp., Prague, Czech Republic
(October 2009).
[10] U. Okoroanyanwu, et al. “Assessing EUV mask defectivity,” Proc. SPIE 7636, (2010).
[11] S. Huh, et al. “Impact of EUV mask absorber defect with pattern-roughness on
lithographic images,” Proc. SPIE 7636, (2010).
[12] S. Lok, “EUV Lithography with the Alpha Demo Tools,” presented at 2009 Int.
SEMATECH EUVL Symp., Prague, Czech Republic (October 2009).
[13] LaFontaine, B., et al., “The use of EUV lithography to produce demonstration devices,”
Proc. SPIE 6921, (2008).
[14] Vandentop, G., et al., “Demonstration of full-field patterning of 32 nm test chips using
EUVL,” Proc. SPIE 7271, (2009).
[15] Naulleau, P. P., et al. “Extreme ultraviolet microexposures at the Advanced Light Source
using the 0.3 numerical aperture micro-exposure tool optic,” J. Vac. Sci. Technol. B 22, 6
(2004).
[16] Uzama, S. et al., “Path to the HVM in EUVL through the development and evaluation of
the SFET,” Proc. SPIE 6517, (2007).
[17] Terasawa, T. et al., “Development of actinic full-field EUV mask blank inspection tool at
MIRAI-Selete,” Proc. SPIE 7271, (2009).
[18] K. A. Goldberg, et al., "Actinic extreme ultraviolet mask inspection beyond 0.25
numerical aperture, "Journal of Vacuum Science and Technology B 26 , 2220-4 (2008).
69
[19] Kinoshita, H. et al., “Aerial Image Mask Inspection System for Extreme Ultraviolet
Lithography,” Jpn. J. Appl. Phys. 46 (2007).
[20] Brizuela, F. et al., “Inspection 13.2-nm table-top full-field microscope”, Proc. SPIE 7271,
(2009).
[21] K. A. Goldberg, I. Mochi, S. Huh , "Collecting EUV mask images through focus by
wavelength tuning," SPIE 7271, (2009).
[22] Liang, T., et al.,” Growth and printability of multilayer phase defects on extreme
ultraviolet mask blanks,” J. Vac. Sci. Technol. B 25, 2098 (2007).
[23] Cho, W., et al., “Detectability and printability of EUVL mask blank defects for the 32 nm
HP node,” Proc. SPIE 6730, (2007).
[24] K. Hamamoto,et al., " Mask defect inspection using an extreme ultraviolet microscope," J.
Vac. Sci. Technol. B 23, 2852–2855 (2005).
[25] Pistor, T., et al., "Extreme ultraviolet mask defect simulation: low-profile defects," J. Vac.
Sci. Technol. B 18, 2926–2929 (2000).
[26] Lam, M.C., et al., “Modeling methodologies and defect printability maps for buried
defects in EUV mask blanks,” Proc. SPIE 6151, (2007).
[27] Smaali, R. et al., “Three-dimensional Rigorous Simulation of EUV Defective Masks
Using Modal Method by Fourier Expansion,” Proc. SPIE 6151, (2006).
[28] Evanschitzky, P., et al. “Fast near field simulation of optical and EUV masks using the
waveguide method,” Proc. of SPIE 6533, (2007).
[29] Burger, S., et al., “Benchmark of FEM, Waveguide and FDTD Algorithms for Rigorous
Mask Simulation,” Proc. SPIE 5992, (2005).
[30] Burger, S. et al., “Benchmark of Rigorous Methods for Electromagnetic Field
Simulations,” Proc. SPIE 7122, (2008).
[31] Gordon, R. L., et al., “Mask Topography Simulation for EUV Lithography,” Proc. SPIE
3676, (1999).
[32] Pistor, T., et al., “Calculating Aerial Images from EUV Masks,” Proc. SPIE 3676 (1999)
[33] Bollepalli, B.S., et al. “On the computation of reflected images from extreme ultraviolet
masks.” Proc. SPIE 3676, (1999).
[34] Evanschitzky, P., et al. “Simulation of Extreme Ultraviolet Masks with Defective
Multilayers,” Proc. SPIE 5130, (2003)
[35] Ito, M., et al., “Simulation of Multilayer Defects in Extreme Ultraviolet Masks,” Jpn. J.
Appl. Phys. Vol. 40, 2549-2553 (2001)
[36] Gullikson, E. M., et al. “Practical approach for modeling extreme ultraviolet lithography
mask defects,” J. Vac. Sci. Technol. B 20, 81 (2002).
[37] Deng, Y., et al., “Models for characterizing the printability of buried EUV defects,” Proc.
SPIE 4343 (2001).
[38] Lam, M.C., et al., “A 3-D substrate and buried defect simulator for EUV mask blanks,”
Proc. SPIE 5751, (2005).
70
[39] Lam, M.C., et al., “Simplified model for absorber feature transmissions on EUV masks,”
Proc. SPIE 6349, (2006).
[40] Clifford, C.H., et al. “Fast Three-Dimensional Simulation of Buried EUV Mask Defect
Interaction with Absorber Features,” Proc. of SPIE 6730, (2007).
[41] Clifford, C. H., et al. “Smoothing Based Model for Images of Isolated Buried EUV
Multilayer Defects,” Proc. of SPIE 6921, (2008).
[42] H. Han, “Characterization of EUV Mask Defects: Printability and Repair Process”
presented at 2008 Int. SEMATECH EUVL Symp., Lake Tahoe (September 2008).
[43] Stearns, D.G., et al., “Localized defects in multilayer coatings”, Thin Solid Films 446
Issue 1, 37-49 (2004).
[44] W. Cho, et al., “Inspection with the Lasertec M7360 at the SEMATECH Mask Blank
Development Center,” Proc. of SPIE 6517, (2007).
[45] S. Stokowski, et al. “Inspecting EUV mask blanks with a 193nm system”, Proc. SPIE
7636, (2010).
[46] Mochi, I. et al., “Improving the performance of the Actinic Inspection Tool with an
optimized alignment procedure,” Proc. of SPIE 7271, (2009). [47] Miura, T, et al, “Nikon EUVL development progress update”, Proc of SPIE Vol 7271
(2009)
[48] Terasawa, T, et al, “High Speed Actinic EUV Mask Blank Inspection with Dark-Field
Imaging,” Proc of SPIE Vol 5446 (2004)
[49] Yang, P, et al, “EUVL ML mask blank fiducial mark application for ML defect
mitigation”, Proc of SPIE Vol 7488 (2009)
[50] T. Liang, “Multilayer Defect Compensation to Enable Quality Masks for EUVL
Production,” presented at 2008 Int. SEMATECH EUVL Symp., Lake Tahoe (September
2008).
[51] S. Huh, et al. “A Study of Real Native Defects on EUV Mask Blanks Using a DUV Blank
Inspection Tool and SEMATECH AIT,” presented at 2009 Int. SEMATECH EUVL
Symp., Prague, Czech Republic (October 2009).